Skip to main content
BenchLM
Data

Terminal-Bench 4.1

We show this table for reference; we do not rank on it.

The current Terminal-Bench release keeps the 66 professional computer-work tasks of 4.0 and regrades every saved trial with verifiers hardened against reward hacking and overly strict checks.

All releases

This page carries every Terminal-Bench release: 4.1 above, then the 4.0, 3.0, 2.1, 2.0 and 1.0 tables and a cross-version ranking. Versions that change the task set or resources are not a like-for-like model comparison.

Jump to version history

Tasks completed on Terminal-Bench 4.1 — November 17, 2026 snapshot

We mirror the published tasks completed view for Terminal-Bench 4.1. Copilot Fusion leads the public snapshot at 71.21%, followed by Claude Opus 5.5 (65.76%) and Claude Fable 5.2 (65.76%). We do not use these results to rank models overall.

63 modelsAgentic9% of Agentic reference weightCurrentUpdated November 17, 2026 snapshot

Current releaseTerminal-Bench 4.1 · Released · 66 tasks · Official leaderboard · Regrades the 4.0 trials

Tasks completed table (63 models)

Tasks completed results for Terminal-Bench 4.1
RankModel / configurationScoreParameters (B)Open / closed
1Copilot FusionGitHubGitHub Copilot · max reasoning
71.21%
Not reportedUnknown
2Claude Opus 5.5AnthropicClaude Code · max reasoning
65.76%
Not reportedClosed
3Claude Fable 5.2AnthropicClaude Code · max reasoning
65.76%
Not reportedClosed
4Gemini 4 CarbonGoogleAntigravity CLI · high reasoning
64.55%
Not reportedClosed
5Claude Sonnet 5.5AnthropicClaude Code · max reasoning
63.94%
Not reportedClosed
6GPT-6.1 SolOpenAICodex · max reasoning
60.91%
Not reportedClosed
7GPT-6 AstraOpenAICodex · max reasoning
60.30%
Not reportedClosed
8GPT-6 AstraOpenAICodex · high reasoning
60.00%
Not reportedClosed
9GPT-6 AstraOpenAICodex · xhigh reasoning
59.70%
Not reportedClosed
10Kimi K3.1 ProMoonshot AIClaude Code · max reasoning
59.70%
Not reportedOpen
11Claude Fable 5.1AnthropicClaude Code · xhigh reasoning
57.58%
Not reportedClosed
12Claude Fable 5.1AnthropicClaude Code · max reasoning
57.27%
Not reportedClosed
13Gemini 4 HydrogenGoogleAntigravity CLI · high reasoning
56.36%
Not reportedClosed
14Gemini 4 ArgonGoogleAntigravity CLI · high reasoning
55.45%
Not reportedClosed
15GPT-6 AstraOpenAICodex · medium reasoning
55.15%
Not reportedClosed
16Claude Fable 5.1AnthropicClaude Code · high reasoning
54.85%
Not reportedClosed
17Claude Fable 5.1AnthropicClaude Code · medium reasoning
53.94%
Not reportedClosed
18Claude Opus 5AnthropicClaude Code · xhigh reasoning
53.33%
Not reportedClosed
19GPT-6.1 TerraOpenAICodex · max reasoning
53.33%
Not reportedClosed
20Claude Opus 5AnthropicClaude Code · max reasoning
52.12%
Not reportedClosed
21GPT-6 AstraOpenAICodex · low reasoning
50.91%
Not reportedClosed
22GPT-6 SolOpenAICodex · max reasoning
50.30%
Not reportedClosed
23Claude Opus 5AnthropicClaude Code · high reasoning
49.70%
Not reportedClosed
24Muse WonderMetaMuse Code · xhigh reasoning
46.97%
Not reportedClosed
25Claude Opus 5AnthropicClaude Code · medium reasoning
45.15%
Not reportedClosed
26Claude Fable 5AnthropicClaude Code · max reasoning
44.85%
Not reportedClosed
27Claude Fable 5.1AnthropicClaude Code · low reasoning
43.33%
Not reportedClosed
28GLM-5.3Z.AIClaude Code · max reasoning
41.21%
Not reportedOpen
29Claude Haiku 5.5AnthropicClaude Code · max reasoning
40.00%
Not reportedClosed
30Grok 4.7xAIGrok Build · xhigh reasoning
38.79%
Not reportedClosed
31GPT-5.6 SolOpenAICodex · max reasoning
38.48%
Not reportedClosed
32DeepSeek V4.1 Flash 1017DeepSeekClaude Code · max reasoning
38.48%
Not reportedOpen
33Hy4TencentClaude Code · max reasoning
37.88%
Not reportedPending
34GLM-5.3-FlashZ.AIClaude Code · max reasoning
36.97%
Not reportedOpen
35GLM-5.3-FlashZ.AIClaude Code · none reasoning
35.45%
Not reportedOpen
36Claude Opus 5AnthropicClaude Code · low reasoning
34.85%
Not reportedClosed
37Muse EmberMetaMuse Code · xhigh reasoning
30.61%
Not reportedClosed
38Mistral Large 4MistralClaude Code · max reasoning
27.58%
Not reportedOpen
39Qwen3.8-Max-0902AlibabaClaude Code · max reasoning
26.97%
Not reportedUnknown
40MiMo-V2.6-ProXiaomimini-swe-agent · none reasoning
26.97%
Not reportedOpen
41Claude Opus 4.8AnthropicClaude Code · max reasoning
23.64%
Not reportedClosed
42Kimi K3Moonshot AIClaude Code · max reasoning
23.03%
Not reportedOpen
43Ember-1Fireworksmini-swe-agent · none reasoning
21.82%
Not reportedClosed
44GPT-5.6 TerraOpenAICodex · max reasoning
21.52%
Not reportedClosed
45Grok 4.6xAIGrok Build · high reasoning
19.70%
Not reportedClosed
46Hy4TencentClaude Code · max reasoning
19.09%
Not reportedPending
47Gemini 3.8 FlashGooglemini-SWE-agent · high reasoning
18.18%
Not reportedClosed
48GPT-5.6 LunaOpenAICodex · max reasoning
17.58%
Not reportedClosed
49MiMo-V2.6-FlashXiaomimini-swe-agent · none reasoning
16.06%
Not reportedOpen
50GPT-6 LunaOpenAICodex · max reasoning
15.15%
Not reportedClosed
51Muse Spark 1.3MetaMuse Code · xhigh reasoning
14.85%
Not reportedClosed
52Inkling Small 2Thinking Machines Labmini-swe-agent · xhigh reasoning
13.03%
Not reportedUnknown
53Gemini 3.7 FlashGooglemini-SWE-agent · high reasoning
12.42%
Not reportedClosed
54Grok 4.5xAIGrok Build · high reasoning
12.42%
Not reportedClosed
55Claude Sonnet 5AnthropicClaude Code · max reasoning
12.12%
Not reportedClosed
56MiniMax M3MiniMaxmini-swe-agent · none reasoning
7.58%
Not reportedOpen
57DeepSeek V4 ProDeepSeekClaude Code · max reasoning
6.67%
Not reportedOpen
58Beam 1.0 501B-A23BReflection AImini-swe-agent · none reasoning
6.36%
Not reportedOpen
59Gemini 3.1 ProGoogleAntigravity CLI · high reasoning
2.73%
Not reportedClosed
60GPT-5.4 miniOpenAICodex · xhigh reasoning
1.82%
Not reportedClosed
61Inkling-SmallThinking Machines Labmini-swe-agent · none reasoning
0.91%
Not reportedOpen
62Nemotron 3.5 Super VL 120B A12BNVIDIAmini-swe-agent · none reasoning
0.30%
Not reportedOpen
63InklingThinking Machines Labmini-swe-agent · none reasoning
0.00%
Not reportedOpen

Results reported by model developersUnranked

Labs ran these themselves. They take no rank here, and they are not comparable to the board score above.

Terminal-Bench 4.1 results reported by model developers, unranked
Gemini 4 HydrogenGoogle57.0%Cheating-adjusted average rewardGoogle's own mini-swe-agent vision harness, five attempts per task, high effortThe stored field follows the official leaderboardNot comparable to the board scoreReported by Google

A lab’s own result is shown for reference. It never takes a rank on this board.

How we label sources

Previous releaseReleased · 66 tasks · Official leaderboard · Not comparable with 3.0 or earlier

Terminal-Bench 4.0

We mirror the official 4.0 table. The raw results compare complete model-and-agent systems. The latest normalized scores replace 3.0 within one external agentic benchmark family and cannot independently make a category ranking-eligible.

Terminal-Bench 4.0; source configurations remain separate from the canonical benchmark table.
Source rankSource model or configurationTasks completedSource setupParameters (B)Open / closed
1Claude Opus 5.5Anthropic64.85%Claude Code · max reasoning · Original resultNot reportedClosed
2Claude Sonnet 5.5Anthropic61.82%Claude Code · max reasoning · Original resultNot reportedClosed
3Kimi K3.1 ProMoonshot AI60.91%Claude Code · max reasoning · Original resultNot reportedOpen
4GPT-6 AstraOpenAI58.18%Codex · max reasoning · Original resultNot reportedClosed
4GPT-6.1 SolOpenAI58.18%Codex · max reasoning · Original resultNot reportedClosed
6Claude Fable 5.1Anthropic57.88%Claude Code · max reasoning · Original resultNot reportedClosed
6GPT-6 AstraOpenAI57.88%Codex · xhigh reasoning · Original resultNot reportedClosed
6GPT-6 AstraOpenAI57.88%Codex · high reasoning · Original resultNot reportedClosed
6Claude Fable 5.1Anthropic57.88%Claude Code · xhigh reasoning · Original resultNot reportedClosed
10Gemini 4 ArgonGoogle54.85%Antigravity CLI · high reasoning · Original resultNot reportedClosed
11Claude Fable 5.1Anthropic54.55%Claude Code · high reasoning · Original resultNot reportedClosed
12GPT-6 AstraOpenAI54.24%Codex · medium reasoning · Original resultNot reportedClosed
13Claude Fable 5.1Anthropic53.94%Claude Code · medium reasoning · Original resultNot reportedClosed
13Claude Opus 5Anthropic53.94%Claude Code · xhigh reasoning · Original resultNot reportedClosed
15Claude Opus 5Anthropic51.82%Claude Code · max reasoning · Original resultNot reportedClosed
15GPT-6.1 TerraOpenAI51.82%Codex · max reasoning · Original resultNot reportedClosed
17GPT-6 AstraOpenAI50.61%Codex · low reasoning · Original resultNot reportedClosed
18Claude Opus 5Anthropic50.30%Claude Code · high reasoning · Original resultNot reportedClosed
19GPT-6 SolOpenAI49.39%Codex · max reasoning · Original resultNot reportedClosed
20Muse WonderMeta46.97%Muse Code · xhigh reasoning · Original resultNot reportedClosed
21Claude Opus 5Anthropic44.85%Claude Code · medium reasoning · Original resultNot reportedClosed
22Claude Fable 5Anthropic44.55%Claude Code · max reasoning · Original resultNot reportedClosed
23Claude Fable 5.1Anthropic43.33%Claude Code · low reasoning · Original resultNot reportedClosed
24GLM-5.3Z.AI41.82%Claude Code · max reasoning · Original resultNot reportedOpen
25DeepSeek V4.1 Flash 1017DeepSeek40.00%Claude Code · max reasoning · Original resultNot reportedOpen
25Claude Haiku 5.5Anthropic40.00%Claude Code · max reasoning · Original resultNot reportedClosed
27Hy4Tencent39.09%Claude Code · max reasoning · Original resultNot reportedPending
28Grok 4.7xAI37.58%Grok Build · xhigh reasoning · Original resultNot reportedClosed
29GPT-5.6 SolOpenAI37.27%Codex · max reasoning · Original resultNot reportedClosed
30GLM-5.3-FlashZ.AI36.67%Claude Code · max reasoning · Original resultNot reportedOpen
31GLM-5.3-FlashZ.AI35.76%Claude Code · none reasoning · Original resultNot reportedOpen
32Claude Opus 5Anthropic34.85%Claude Code · low reasoning · Original resultNot reportedClosed
33Mistral Large 4Mistral27.88%Claude Code · max reasoning · Original resultNot reportedOpen
34Qwen3.8-Max-0902Alibaba26.97%Claude Code · max reasoning · Original resultNot reportedUnknown
34MiMo-V2.6-ProXiaomi26.97%mini-swe-agent · none reasoning · Original resultNot reportedOpen
36Claude Opus 4.8Anthropic23.64%Claude Code · max reasoning · Original resultNot reportedClosed
37Kimi K3Moonshot AI22.42%Claude Code · max reasoning · Original resultNot reportedOpen
38GPT-5.6 TerraOpenAI21.52%Codex · max reasoning · Original resultNot reportedClosed
39Grok 4.6xAI20.30%Grok Build · high reasoning · Original resultNot reportedClosed
40Hy4Tencent19.39%Claude Code · max reasoning · Original resultNot reportedPending
41Gemini 3.8 FlashGoogle19.09%mini-SWE-agent · high reasoning · Original resultNot reportedClosed
42GPT-5.6 LunaOpenAI17.27%Codex · max reasoning · Original resultNot reportedClosed
43GPT-6 LunaOpenAI16.36%Codex · max reasoning · Original resultNot reportedClosed
44MiMo-V2.6-FlashXiaomi16.06%mini-swe-agent · none reasoning · Original resultNot reportedOpen
45Muse Spark 1.3Meta14.55%Muse Code · xhigh reasoning · Original resultNot reportedClosed
46Grok 4.5xAI12.42%Grok Build · high reasoning · Original resultNot reportedClosed
46Claude Sonnet 5Anthropic12.42%Claude Code · max reasoning · Original resultNot reportedClosed
48Gemini 3.7 FlashGoogle11.21%mini-SWE-agent · high reasoning · Original resultNot reportedClosed
49MiniMax M3MiniMax7.27%mini-swe-agent · none reasoning · Original resultNot reportedOpen
50DeepSeek V4 ProDeepSeek6.67%Claude Code · max reasoning · Original resultNot reportedOpen
51Beam 1.0 501B-A23BReflection AI6.36%mini-swe-agent · none reasoning · Original resultNot reportedOpen
52Gemini 3.1 ProGoogle2.73%Antigravity CLI · high reasoning · Original resultNot reportedClosed
53GPT-5.4 miniOpenAI1.82%Codex · xhigh reasoning · Original resultNot reportedClosed
54Inkling-SmallThinking Machines Lab0.91%mini-swe-agent · none reasoning · Original resultNot reportedOpen
55Nemotron 3.5 Super VL 120B A12BNVIDIA0.30%mini-swe-agent · none reasoning · Original resultNot reportedOpen
56InklingThinking Machines Lab0.00%mini-swe-agent · none reasoning · Original resultNot reportedOpen

AA Terminal-Bench 4.0

Stored separately from the benchmark-owned Terminal-Bench 4.0 lane because Artificial Analysis controls the agent harness and configuration. Display-only; not yet admitted as independent-run evidence.

AA Terminal-Bench 4.0; source configurations remain separate from the canonical benchmark table.
Source rankSource model or configurationScoreWho ran itSource setupParameters (B)Open / closed
1Kimi K3.1 ProMoonshot AI67.2%Carried from an upstream pageArtificial Analysis · Original resultNot reportedOpen
2Gemini 4 CarbonGoogle65.2%Reported by GoogleArtificial Analysis · Original resultNot reportedClosed
3GPT-6.2 AstraOpenAI64.1%Carried from an upstream pageArtificial Analysis · Original resultNot reportedClosed
4Claude Sonnet 5.5Anthropic63.6%Carried from an upstream pageArtificial Analysis · Original resultNot reportedClosed
5GPT-6.2 SolOpenAI60.1%Carried from an upstream pageArtificial Analysis · Original resultNot reportedClosed
6Claude Fable 5.2Anthropic59.6%Carried from an upstream pageArtificial Analysis · Original resultNot reportedClosed
7Claude Opus 5.5Anthropic59.6%Carried from an upstream pageArtificial Analysis · Original resultNot reportedClosed
8Muse Wonder FullMeta57.1%Reported by MetaArtificial Analysis · Original resultNot reportedClosed
9GPT-6.1 TerraOpenAI55.6%Carried from an upstream pageArtificial Analysis · Original resultNot reportedClosed
10Gemini 4 HydrogenGoogle53.5%Carried from an upstream pageArtificial Analysis · Original resultNot reportedClosed
11Muse WonderMeta52.0%Reported by MetaArtificial Analysis · Original resultNot reportedClosed
12Qwen4 MaxAlibaba48.5%Carried from an upstream pageArtificial Analysis · Original resultNot reportedClosed
13GLM-5.3Z.AI41.9%Carried from an upstream pageArtificial Analysis · Original resultNot reportedOpen
14Muse EmberMeta40.9%Carried from an upstream pageArtificial Analysis · Original resultNot reportedClosed
15MiMo-V2.6-ProXiaomi34.8%Carried from an upstream pageArtificial Analysis · Original resultNot reportedOpen
16Claude Haiku 5.5Anthropic32.8%Carried from an upstream pageArtificial Analysis · Original resultNot reportedClosed
17Grok 4.7xAI25.8%Carried from an upstream pageArtificial Analysis · Original resultNot reportedClosed
18Mistral Large 4Mistral24.7%Carried from an upstream pageArtificial Analysis · Original resultNot reportedOpen
19Inkling 2 SmallThinking Machines Lab13.1%Carried from an upstream pageArtificial Analysis · Original resultNot reportedClosed
20GPT-6 LunaOpenAI12.6%Carried from an upstream pageArtificial Analysis · Original resultNot reportedClosed
21Qwen3.8-27BAlibaba5.6%Carried from an upstream pageArtificial Analysis · Original resultNot reportedOpen
22Beam 1.0 501B-A23BReflection AI3.0%Carried from an upstream pageArtificial Analysis · Original resultNot reportedOpen

Terminal-Bench 4.0 (Vals)

We keep the Vals-hosted Terminal-Bench 4.0 table on a dedicated key so it never overwrites the official Terminal-Bench 4.0 leaderboard rows. Vals runs its own harness and publishes per-domain splits; this mirror is display only and does not enter weighted rankings.

Terminal-Bench 4.0 (Vals); source configurations remain separate from the canonical benchmark table.
Source rankSource model or configurationVals Terminal-Bench 4.0 mirror scoreSource setupParameters (B)Open / closed
1Claude Fable 5.2Anthropic65.15%Vals AI · Original resultNot reportedClosed
2Claude Opus 5.5Anthropic65.15%Vals AI · Original resultNot reportedClosed
3Claude Sonnet 5.5Anthropic64.14%Vals AI · Original resultNot reportedClosed
4Kimi K3.1 ProMoonshot AI63.64%max reasoning · Original resultNot reportedOpen
5GPT-6.2 AstraOpenAI63.13%max reasoning · Original resultNot reportedClosed
6Gemini 4 CarbonGoogle63.13%high reasoning · Original resultNot reportedClosed
7GPT-6.2 SolOpenAI60.61%max reasoning · Original resultNot reportedClosed
8GPT-6 AstraOpenAI59.60%max reasoning · Original resultNot reportedClosed
9Claude Fable 5.1Anthropic58.08%Vals AI · Original resultNot reportedClosed
10Muse Wonder FullMeta58.08%max reasoning · Original resultNot reportedClosed
11Gemini 4 ArgonGoogle57.58%high reasoning · Original resultNot reportedClosed
12GPT-6.1 SolOpenAI55.05%max reasoning · Original resultNot reportedClosed
13Gemini 4 HydrogenGoogle54.04%xhigh reasoning · Original resultNot reportedClosed
14Claude Opus 5Anthropic53.53%Vals AI · Original resultNot reportedClosed
15GPT-6.1 TerraOpenAI49.49%max reasoning · Original resultNot reportedClosed
16Muse WonderMeta49.49%max reasoning · Original resultNot reportedClosed
17Qwen4 MaxAlibaba48.48%xhigh reasoning · Original resultNot reportedClosed
18GPT-6 SolOpenAI44.44%max reasoning · Original resultNot reportedClosed
19Claude Fable 5Anthropic41.41%Vals AI · Original resultNot reportedClosed
20Muse EmberMeta39.39%max reasoning · Original resultNot reportedClosed
21Hy4Tencent39.39%Vals AI · Original resultNot reportedPending
22GLM-5.3Z.AI38.89%max reasoning · Original resultNot reportedOpen
23Qwen3.8 Max InternalAlibaba38.89%Vals AI · Original resultNot reportedUnknown
24GPT-5.6 SolOpenAI37.88%max reasoning · Original resultNot reportedClosed
25GLM-5.3-FlashZ.AI37.37%max reasoning · Original resultNot reportedOpen
26Claude Haiku 5.5Anthropic36.36%Vals AI · Original resultNot reportedClosed
27Qwen3.8 MaxAlibaba34.34%Vals AI · Original resultNot reportedOpen
28DeepSeek V4.1 Flash 1017DeepSeek31.82%max reasoning · Original resultNot reportedOpen
29Step 5 PreviewStepFun31.82%Vals AI · Original resultNot reportedOpen
30MiMo-V2.6-ProXiaomi31.31%Vals AI · Original resultNot reportedOpen
31Grok 4.7xAI28.79%xhigh reasoning · Original resultNot reportedClosed
32Mistral Large 4Mistral27.78%high reasoning · Original resultNot reportedOpen
33Mistral Large 4 OssMistral AI26.26%high reasoning · Original resultNot reportedUnknown
34MiniMax M3.1 FlashMiniMax25.25%Vals AI · Original resultNot reportedClosed
35Muse Spark 1.3 MaxMeta24.75%max reasoning · Original resultNot reportedUnknown
36MiMo-V2.6-FlashXiaomi24.24%Vals AI · Original resultNot reportedOpen
37Claude Opus 4.8Anthropic23.23%Vals AI · Original resultNot reportedClosed
38GPT-5.6 TerraOpenAI22.73%max reasoning · Original resultNot reportedClosed
39Ember-1Fireworks20.71%Vals AI · Original resultNot reportedClosed
40DeepSeek V4.1 Flash 1017DeepSeek19.70%high reasoning · Original resultNot reportedOpen
41Gemini 3.8 FlashGoogle19.70%high reasoning · Original resultNot reportedClosed
42DeepSeek V4 Flash 0731DeepSeek18.69%high reasoning · Original resultNot reportedOpen
43Grok 4.6xAI17.17%high reasoning · Original resultNot reportedClosed
44Kimi K3Moonshot AI17.17%max reasoning · Original resultNot reportedOpen
45Hy4 flashTencent15.15%Vals AI · Original resultNot reportedPending
46DeepSeek V4 Pro 0813DeepSeek14.14%max reasoning · Original resultNot reportedUnknown
47GPT-6 LunaOpenAI13.64%max reasoning · Original resultNot reportedClosed
48Inkling 2 SmallThinking Machines Lab12.63%0.99 reasoning · Original resultNot reportedClosed
49Gemini 3.7 FlashGoogle12.12%high reasoning · Original resultNot reportedClosed
50GPT-5.6 LunaOpenAI11.62%max reasoning · Original resultNot reportedClosed
51DeepSeek V4 ProDeepSeek11.11%max reasoning · Original resultNot reportedOpen
52Claude Sonnet 5Anthropic10.61%Vals AI · Original resultNot reportedClosed
53Muse Spark 1.3Meta10.61%xhigh reasoning · Original resultNot reportedClosed
54Grok 4.5xAI8.59%high reasoning · Original resultNot reportedClosed
55Hy4 previewTencent8.08%Vals AI · Original resultNot reportedOpen
56Muse Spark OpenMeta6.57%xhigh reasoning · Original resultNot reportedOpen
57Gemini 3.5 FlashGoogle6.06%high reasoning · Original resultNot reportedClosed
58Muse Spark 1.2Meta6.06%xhigh reasoning · Original resultNot reportedClosed
59Beam 1.0Reflection4.04%Vals AI · Original resultNot reportedUnknown
60Mercury 3Inception3.03%high reasoning · Original resultNot reportedClosed
61Gemini 3.1 Pro PreviewGoogle2.52%high reasoning · Original resultNot reportedUnknown
62GPT-5.4 miniOpenAI2.52%xhigh reasoning · Original resultNot reportedClosed
63MiniMax M3MiniMax1.01%Vals AI · Original resultNot reportedOpen
64InklingThinking Machines Lab0.51%0.99 reasoning · Original resultNot reportedOpen
65MiMo-V2.5-ProXiaomi0.51%Vals AI · Original resultNot reportedClosed
66Inkling-SmallThinking Machines Lab0.00%0.99 reasoning · Original resultNot reportedOpen
67Nemotron 3.5 Super VL 120B A12BNVIDIA0.00%high reasoning · Original resultNot reportedOpen
68Mercury 2.5Inception0.00%high reasoning · Original resultNot reportedClosed

Terminal-Bench version history

Every Terminal-Bench release, newest first. Scores are not directly comparable across releases that change the task set or resources (1.0, 2.x, 3.0 and 4.0). Version 4.1 is a regrade of the 4.0 trials, so its scores stay within each entry’s 4.0 confidence interval.

Terminal-Bench releases with their release date, task set, change and published leader.
VersionReleasedTasksWhat changedLeaderModels
Terminal-Bench 4.1CurrentNov 9, 202666 tasksRegrades the saved 4.0 trials with verifiers hardened against reward hacking; no agent-facing changes.LeaderCopilot Fusion 71.2%Official board, GitHub Copilot49 models
Terminal-Bench 4.0Aug 28, 202666 tasksRecalibrates task resources, fixes 19 tasks and removes 8; fresh trials, not comparable with 3.0.LeaderClaude Opus 5.5 64.8%Official board, Claude Code42 models
Terminal-Bench 3.0Jul 30, 202674 tasks, 7 domainsNew professional computer-work task set, first published as FrontierBench.LeaderClaude Opus 5 42.7%Official board, mini-SWE-agent12 models
Terminal-Bench 2.1May 6, 202689 tasksFixes 28 Terminal-Bench 2.0 tasks. No official leaderboard; results are provider reports.LeaderSWE-2 92.8%Highest provider report89 models
Terminal-Bench 2.0Nov 7, 202589 tasksHarder terminal task set released with the Harbor evaluation framework.LeaderGPT-5.5 84.7%Official board, NexAU-AHE48 models
Terminal-Bench 1.0May 19, 202580 tasks at launchThe first Terminal-Bench task set (Terminal-Bench-Core). The table is the Vals AI run, split into easy, medium and hard tasks.LeaderGPT-5.2 63.8%Highest score in the Vals AI run47 models

Rankings across Terminal-Bench versions

Each cell is the model’s rank among models on that release’s official leaderboard, using its best published agent and effort setting, with the score below. Rows cover every model on the 4.1, 4.0 or 3.0 boards. Versions 2.1 and 1.0 have no official leaderboard here, so they have no column. A dash means the model has no row on that board.

Model rank and score on each official Terminal-Bench leaderboard.
ModelTerminal-Bench 4.149 modelsTerminal-Bench 4.042 modelsTerminal-Bench 3.012 modelsTerminal-Bench 2.048 models
Copilot Fusion#171.2%Not on this boardNot on this boardNot on this board
Claude Opus 5.51 (Down 1 place from 4.0)#265.8%#164.8%Not on this boardNot on this board
Claude Fable 5.2#265.8%Not on this boardNot on this boardNot on this board
Gemini 4 Carbon#464.5%Not on this boardNot on this boardNot on this board
Claude Sonnet 5.53 (Down 3 places from 4.0)#563.9%#261.8%Not on this boardNot on this board
GPT-6.1 Sol2 (Down 2 places from 4.0)#660.9%#458.2%Not on this boardNot on this board
GPT-6 Astra3 (Down 3 places from 4.0)#760.3%#458.2%Not on this boardNot on this board
Kimi K3.1 Pro5 (Down 5 places from 4.0)#859.7%#360.9%Not on this boardNot on this board
Claude Fable 5.13 (Down 3 places from 4.0)#957.6%#657.9%Not on this boardNot on this board
Gemini 4 Hydrogen#1056.4%Not on this boardNot on this boardNot on this board
Gemini 4 Argon4 (Down 4 places from 4.0)#1155.5%#754.9%Not on this boardNot on this board
Claude Opus 54 (Down 4 places from 4.0)#1253.3%#853.9%#142.7%Not on this board
GPT-6.1 Terra3 (Down 3 places from 4.0)#1253.3%#951.8%Not on this boardNot on this board
GPT-6 Sol4 (Down 4 places from 4.0)#1450.3%#1049.4%Not on this boardNot on this board
Muse Wonder4 (Down 4 places from 4.0)#1547.0%#1147.0%Not on this boardNot on this board
Claude Fable 54 (Down 4 places from 4.0)#1644.9%#1244.5%#334.0%Not on this board
GLM-5.34 (Down 4 places from 4.0)#1741.2%#1341.8%#432.4%Not on this board
Claude Haiku 5.54 (Down 4 places from 4.0)#1840%#1440%Not on this boardNot on this board
Grok 4.72 (Down 2 places from 4.0)#1938.8%#1737.6%Not on this boardNot on this board
DeepSeek V4.1 Flash 10176 (Down 6 places from 4.0)#2038.5%#1440%Not on this boardNot on this board
GPT-5.6 Sol2 (Down 2 places from 4.0)#2038.5%#1837.3%#234.6%Not on this board
Hy46 (Down 6 places from 4.0)#2237.9%#1639.1%Not on this boardNot on this board
GLM-5.3-Flash4 (Down 4 places from 4.0)#2337.0%#1936.7%Not on this boardNot on this board
Muse Ember#2430.6%Not on this boardNot on this boardNot on this board
Mistral Large 45 (Down 5 places from 4.0)#2527.6%#2027.9%Not on this boardNot on this board
MiMo-V2.6-Pro5 (Down 5 places from 4.0)#2627.0%#2127.0%Not on this boardNot on this board
Qwen3.8-Max-09025 (Down 5 places from 4.0)#2627.0%#2127.0%Not on this boardNot on this board
Claude Opus 4.85 (Down 5 places from 4.0)#2823.6%#2323.6%#621.1%Not on this board
Kimi K35 (Down 5 places from 4.0)#2923.0%#2422.4%Not on this boardNot on this board
Ember-1#3021.8%Not on this boardNot on this boardNot on this board
GPT-5.6 Terra6 (Down 6 places from 4.0)#3121.5%#2521.5%#720.8%Not on this board
Grok 4.66 (Down 6 places from 4.0)#3219.7%#2620.3%#526.5%Not on this board
Gemini 3.8 Flash6 (Down 6 places from 4.0)#3318.2%#2719.1%Not on this boardNot on this board
GPT-5.6 Luna6 (Down 6 places from 4.0)#3417.6%#2817.3%#1114.3%Not on this board
MiMo-V2.6-Flash5 (Down 5 places from 4.0)#3516.1%#3016.1%Not on this boardNot on this board
GPT-6 Luna7 (Down 7 places from 4.0)#3615.2%#2916.4%Not on this boardNot on this board
Muse Spark 1.36 (Down 6 places from 4.0)#3714.8%#3114.6%Not on this boardNot on this board
Inkling Small 2#3813.0%Not on this boardNot on this boardNot on this board
Grok 4.57 (Down 7 places from 4.0)#3912.4%#3212.4%#915.7%Not on this board
Gemini 3.7 Flash5 (Down 5 places from 4.0)#3912.4%#3411.2%Not on this boardNot on this board
Claude Sonnet 59 (Down 9 places from 4.0)#4112.1%#3212.4%#1014.6%Not on this board
MiniMax M37 (Down 7 places from 4.0)#427.6%#357.3%Not on this boardNot on this board
DeepSeek V4 Pro7 (Down 7 places from 4.0)#436.7%#366.7%Not on this boardNot on this board
Beam 1.0 501B-A23B7 (Down 7 places from 4.0)#446.4%#376.4%Not on this boardNot on this board
Gemini 3.1 Pro7 (Down 7 places from 4.0)#452.7%#382.7%Not on this board#380.2%
GPT-5.4 mini7 (Down 7 places from 4.0)#461.8%#391.8%Not on this boardNot on this board
Inkling-Small7 (Down 7 places from 4.0)#470.9%#400.9%Not on this boardNot on this board
Nemotron 3.5 Super VL 120B A12B7 (Down 7 places from 4.0)#480.3%#410.3%Not on this boardNot on this board
Inkling7 (Down 7 places from 4.0)#490%#420%Not on this boardNot on this board
SWE-1.7 LightningNot on this boardNot on this board#818.6%Not on this board
GLM-5.2Not on this boardNot on this board#124.6%Not on this board

Places gained or lost from 4.0 to 4.1. Version 4.1 regrades the 4.0 trials, so this is the one like-for-like rank change. 42 of 42 models moved.

Older releaseReleased · 74 tasks, 7 domains · Official leaderboard · Not comparable with other releases

Terminal-Bench 3.0

We mirror the official version 0.1 table. The raw results compare complete model-and-agent systems. Their normalized scores count as one external agentic benchmark family and cannot independently make a category ranking-eligible.

Terminal-Bench 3.0; source configurations remain separate from the canonical benchmark table.
Source rankSource model or configurationTasks completedSource setupParameters (B)Open / closed
1Claude Opus 5Anthropic42.7%max reasoning · Original resultNot reportedClosed
2GPT-5.6 SolOpenAI34.6%max reasoning · Original resultNot reportedClosed
3Claude Fable 5Anthropic34.0%max reasoning · Original resultNot reportedClosed
4GLM-5.3Z.AI32.4%max reasoning · Original resultNot reportedOpen
5Grok 4.6xAI26.5%high reasoning · Original resultNot reportedClosed
6Claude Opus 4.8Anthropic21.1%max reasoning · Original resultNot reportedClosed
7GPT-5.6 TerraOpenAI20.8%max reasoning · Original resultNot reportedClosed
8SWE-1.7 LightningCognition18.6%Ryan Marten, Alex Shaw, Andy Konwinski, Harbor, and the Laude Institute · Original resultNot reportedUnknown
9Grok 4.5xAI15.7%xhigh reasoning · Original resultNot reportedClosed
10Claude Sonnet 5Anthropic14.6%max reasoning · Original resultNot reportedClosed
11GPT-5.6 LunaOpenAI14.3%max reasoning · Original resultNot reportedClosed
12GLM-5.2Z.AI4.6%max reasoning · Original resultNot reportedOpen

Older releaseReleased · 89 tasks · Provider reports · Not comparable with other releases

Terminal-Bench 2.1

BenchLM stores Terminal-Bench 2.1 results on their own key, apart from Terminal-Bench 2.0, because the two versions are not directly comparable. Harness and effort settings vary by row.

Terminal-Bench 2.1; source configurations remain separate from the canonical benchmark table.
Source rankSource model or configurationScoreWho ran itSource setupParameters (B)Open / closed
1SWE-2Cognition92.8%Reported by CognitionTerminal-Bench contributors · Original resultNot reportedClosed
2GPT-5.6 SolOpenAI91.9%Reported by OpenAITerminal-Bench contributors · Original resultNot reportedClosed
3DeepSeek V4.1 Flash 1017DeepSeek90.6%Reported by DeepSeekTerminal-Bench contributors · Original resultNot reportedOpen
4MiMo-V2.6-ProXiaomi89.9%Reported by XiaomiTerminal-Bench contributors · Original resultNot reportedOpen
5Gemini 3.8 FlashGoogle89.4%Reported by GoogleTerminal-Bench contributors · Original resultNot reportedClosed
6SWE-2 FlashCognition89.2%Reported by CognitionTerminal-Bench contributors · Original resultNot reportedClosed
7Muse Spark 1.3Meta88.8%Reported by MetaTerminal-Bench contributors · Original resultNot reportedClosed
8LongCat-2.5Meituan88.4%Reported by MeituanTerminal-Bench contributors · Original resultNot reportedOpen
9Kimi K3Moonshot AI88.3%Reported by Moonshot AITerminal-Bench contributors · Original resultNot reportedOpen
10GLM-5.3Z.AI88.2%Reported by Z.AITerminal-Bench contributors · Original resultNot reportedOpen
11Claude Mythos 5Anthropic88.0%Reported by AnthropicTerminal-Bench contributors · Original resultNot reportedClosed
12DeepSeek V4 ProDeepSeek87.9%Reported by DeepSeekTerminal-Bench contributors · Original resultNot reportedOpen
13MiMo-V2.6-FlashXiaomi87.6%Reported by XiaomiTerminal-Bench contributors · Original resultNot reportedOpen
14GPT-5.6 TerraOpenAI87.4%Reported by OpenAITerminal-Bench contributors · Original resultNot reportedClosed
15Qwen3.8 MaxAlibaba86.6%Reported by AlibabaTerminal-Bench contributors · Original resultNot reportedOpen
16Ornith-1.5-397BOrnith AI86.1%Reported by Ornith AITerminal-Bench contributors · Original resultNot reportedOpen
17Gemini 3.7 FlashGoogle85.8%Reported by GoogleTerminal-Bench contributors · Original resultNot reportedClosed
18Hy4 previewTencent85.4%Reported by TencentTerminal-Bench contributors · Original resultNot reportedOpen
19Inkling 2 SmallThinking Machines Lab85.1%Reported by Thinking Machines LabTerminal-Bench contributors · Original resultNot reportedClosed
20Step 5 PreviewStepFun85.0%Reported by StepFunTerminal-Bench contributors · Original resultNot reportedOpen
21GPT-5.6 LunaOpenAI84.7%Reported by OpenAITerminal-Bench contributors · Original resultNot reportedClosed
22MAI Code 2Microsoft84.6%Reported by MicrosoftTerminal-Bench contributors · Original resultNot reportedClosed
23Claude Fable 5Anthropic84.3%Reported by AnthropicTerminal-Bench contributors · Original resultNot reportedClosed
24GLM-5.3-FlashZ.AI84.3%Reported by Z.AITerminal-Bench contributors · Original resultNot reportedOpen
25Grok 4.5xAI83.3%Reported by xAITerminal-Bench contributors · Original resultNot reportedClosed
26Ultramarine 1 FlashPaper Instruments83.2%Reported by Paper InstrumentsTerminal-Bench contributors · Original resultNot reportedClosed
27Muse Spark 1.2Meta82.9%Reported by MetaTerminal-Bench contributors · Original resultNot reportedClosed
28DeepSeek V4 Flash 0731DeepSeek82.7%Reported by DeepSeekTerminal-Bench contributors · Original resultNot reportedOpen
29Sakana Fugu-UltraSakana AI82.1%Reported by Sakana AITerminal-Bench contributors · Original resultNot reportedClosed
30Ember-1Fireworks82.0%Reported by FireworksTerminal-Bench contributors · Original resultNot reportedClosed
31SWE-1.7Cognition81.5%Reported by CognitionTerminal-Bench contributors · Original resultNot reportedClosed
32Ling 3.1 FlashInclusionAI81.2%Reported by InclusionAITerminal-Bench contributors · Original resultNot reportedOpen
33GLM-5.2Z.AI81.0%Reported by Z.AITerminal-Bench contributors · Original resultNot reportedOpen
34Ripple-65B PreviewRicursive Intelligence80.5%Reported by Ricursive IntelligenceTerminal-Bench contributors · Original resultNot reportedOpen
35Claude Sonnet 5Anthropic80.4%Reported by AnthropicTerminal-Bench contributors · Original resultNot reportedClosed
36Sakana FuguSakana AI80.2%Reported by Sakana AITerminal-Bench contributors · Original resultNot reportedClosed
37Beam 1.0 501B-A23BReflection AI80.1%Reported by Reflection AITerminal-Bench contributors · Original resultNot reportedOpen
38Muse Spark 1.1Meta80.0%Reported by MetaTerminal-Bench contributors · Original resultNot reportedClosed
39Particle-1Particle AI79.9%Reported by Particle AITerminal-Bench contributors · Original resultNot reportedClosed
40Beam 1.0 501B-A23B FP8Reflection AI78.9%Reported by Reflection AITerminal-Bench contributors · Original resultNot reportedOpen
41Beam 1.0 501B-A23B NVFP4Reflection AI78.4%Reported by Reflection AITerminal-Bench contributors · Original resultNot reportedOpen
42Atria Dawn PreviewShanghai Artificial Intelligence Laboratory78.3%Reported by the labTerminal-Bench contributors · Original resultNot reportedOpen
43MAI Code 2 FlashMicrosoft78.3%Reported by MicrosoftTerminal-Bench contributors · Original resultNot reportedClosed
44Ornith-1.0-397BDeepReinforce AI77.5%Reported by DeepReinforce AITerminal-Bench contributors · Original resultNot reportedOpen
45Gemini 3.5 FlashGoogle76.2%Reported by GoogleTerminal-Bench contributors · Original resultNot reportedClosed
46dots3-note PreviewDots Studio75.1%Reported by Dots StudioTerminal-Bench contributors · Original resultNot reportedOpen
47Claude Opus 4.8Anthropic74.6%Reported by AnthropicTerminal-Bench contributors · Original resultNot reportedClosed
48Jais-3-375B-A23BIFM73.8%Reported by IFMTerminal-Bench contributors · Original resultNot reportedOpen
49Qwen3.8-27BAlibaba73.0%Reported by AlibabaTerminal-Bench contributors · Original resultNot reportedOpen
50OrcaCyber-Nano-1.0OrcaRouter72.5%Reported by OrcaRouterTerminal-Bench contributors · Original resultNot reportedOpen
51Seed 2.1 ProByteDance71.0%Reported by ByteDanceTerminal-Bench contributors · Original resultNot reportedClosed
52Apodex 1.1Apodex70.8%Reported by ApodexTerminal-Bench contributors · Original resultNot reportedClosed
53Laguna S 2.1Poolside70.2%Reported by PoolsideTerminal-Bench contributors · Original resultNot reportedOpen
54Quasar 438BMultiverse Computing69.3%Reported by Multiverse ComputingTerminal-Bench contributors · Original resultNot reportedClosed
55Ornith-1.5-35B-A3BOrnith AI67.8%Reported by Ornith AITerminal-Bench contributors · Original resultNot reportedOpen
56Seed 2.1 TurboByteDance67.6%Reported by ByteDanceTerminal-Bench contributors · Original resultNot reportedClosed
57Jais-3-MoVA-36B-A4BIFM67.3%Reported by IFMTerminal-Bench contributors · Original resultNot reportedOpen
58Nemotron 3.5 Super VL 120B A12BNVIDIA66.7%Reported by NVIDIATerminal-Bench contributors · Original resultNot reportedOpen
59MiniMax M3MiniMax66.0%Reported by MiniMaxTerminal-Bench contributors · Original resultNot reportedOpen
60Pokee-Isaac 28BPokee AI65.1%Reported by Pokee AITerminal-Bench contributors · Original resultNot reportedClosed
61Inkling-SmallThinking Machines Lab64.7%Reported by Thinking Machines LabTerminal-Bench contributors · Original resultNot reportedOpen
62Ornith-1.0-35BDeepReinforce AI64.2%Reported by DeepReinforce AITerminal-Bench contributors · Original resultNot reportedOpen
63InklingThinking Machines Lab63.8%Reported by Thinking Machines LabTerminal-Bench contributors · Original resultNot reportedOpen
64Jais-3-32BIFM63.7%Reported by IFMTerminal-Bench contributors · Original resultNot reportedOpen
65MAI-Code-1.1-FlashMicrosoft62.9%Reported by MicrosoftTerminal-Bench contributors · Original resultNot reportedClosed
66Nemotron 3.5 Super VL 120B A12B NVFP4NVIDIA60.8%Reported by NVIDIATerminal-Bench contributors · Original resultNot reportedOpen
67Step 3.7 FlashStepFun59.5%Reported by StepFunTerminal-Bench contributors · Original resultNot reportedOpen
68Ling 3.0 FlashInclusionAI57.0%Reported by InclusionAITerminal-Bench contributors · Original resultNot reportedOpen
69Solar Pro 4Upstage57.0%Reported by UpstageTerminal-Bench contributors · Original resultNot reportedClosed
70Nemotron 3 UltraNVIDIA56.4%Reported by NVIDIATerminal-Bench contributors · Original resultNot reportedOpen
71FlintRecursive54.9%Reported by RecursiveTerminal-Bench contributors · Original resultNot reportedOpen
72Gemini 3.5 Flash-LiteGoogle54.0%Reported by GoogleTerminal-Bench contributors · Original resultNot reportedClosed
73Nemotron 3.5 Bolt 74B-A9B BF16NVIDIA53.9%Reported by NVIDIATerminal-Bench contributors · Original resultNot reportedOpen
74Ternary Bonsai 2 27BPrism ML52.8%Reported by Prism MLTerminal-Bench contributors · Original resultNot reportedOpen
75Muse Glimmer 30BMeta51.7%Reported by MetaTerminal-Bench contributors · Original resultNot reportedOpen
76Nemotron 3.5 Bolt 74B-A9BNVIDIA51.0%Reported by NVIDIATerminal-Bench contributors · Original resultNot reportedOpen
77Jais-3-7BIFM46.4%Reported by IFMTerminal-Bench contributors · Original resultNot reportedOpen
78Ornith-1.5-9BOrnith AI46.2%Reported by Ornith AITerminal-Bench contributors · Original resultNot reportedOpen
79K-EXAONE 2.0LG AI Research43.8%Reported by LG AI ResearchTerminal-Bench contributors · Original resultNot reportedOpen
80Ornith-1.0-9BDeepReinforce AI43.1%Reported by DeepReinforce AITerminal-Bench contributors · Original resultNot reportedOpen
81Jais-3-3.7BIFM41.2%Reported by IFMTerminal-Bench contributors · Original resultNot reportedOpen
82A.X K2SK Telecom36.0%Reported by SK TelecomTerminal-Bench contributors · Original resultNot reportedOpen
83Granite 4.2 30BIBM29.2%Reported by IBMTerminal-Bench contributors · Original resultNot reportedOpen
84Mercury 2Inception27.0%Carried from an upstream pageTerminal-Bench contributors · Original resultNot reportedClosed
85Nemotron 3.5 Lightning 30B A3B NVFP4NVIDIA23.5%Reported by NVIDIATerminal-Bench contributors · Original resultNot reportedOpen
86Granite 4.2 8BIBM20.6%Reported by IBMTerminal-Bench contributors · Original resultNot reportedOpen
87Mellum2.1-12B-A2.5B-ThinkingJetBrains17.4%Reported by JetBrainsTerminal-Bench contributors · Original resultNot reportedOpen
88LFM3-1.7BLiquidAI12.9%Reported by LiquidAITerminal-Bench contributors · Original resultNot reportedOpen
89MiniCPM5-2BOpenBMB8.6%Reported by OpenBMBTerminal-Bench contributors · Original resultNot reportedOpen

AA Terminal-Bench 2.1

Stored separately from the core Terminal-Bench lane because Artificial Analysis controls the evaluation configuration.

AA Terminal-Bench 2.1; source configurations remain separate from the canonical benchmark table.
Source rankSource model or configurationScoreWho ran itSource setupParameters (B)Open / closed
1Claude Fable 5.1Anthropic91.4%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedClosed
2Claude Opus 5Anthropic89.1%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedClosed
3Grok 4.6xAI88.4%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedClosed
4GPT-5.6 TerraOpenAI88.0%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedClosed
5Gemini 3.8 FlashGoogle87.6%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedClosed
6Qwen3.8-Flash-NextAlibaba86.1%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedOpen
7Gemini 3.7 FlashGoogle85.8%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedClosed
8Kimi K3Moonshot AI85.0%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedOpen
9Claude Fable 5Anthropic84.6%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedClosed
10Claude Opus 4.8Anthropic84.6%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedClosed
11GPT-5.5OpenAI84.3%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedClosed
12GLM-5.3-FlashZ.AI84.3%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedOpen
13GLM-5.3Z.AI83.9%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedOpen
14Claude Opus 4.7Anthropic83.1%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedClosed
15Grok 4.5xAI81.6%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedClosed
16Inkling 2 SmallThinking Machines Lab81.3%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedClosed
17GPT-5.6 LunaOpenAI80.9%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedClosed
18Claude Sonnet 5Anthropic80.5%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedClosed
19Muse Spark 1.2Meta80.1%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedClosed
20Qwen3.8-27BAlibaba79.8%Independent run: Artificial AnalysisArtificial Analysis · Original resultNot reportedOpen

Terminal-Bench 2.1 (Vals)

BenchLM mirrors this Vals AI leaderboard for reference only. It is excluded from BenchLM overall and category rankings.

Terminal-Bench 2.1 (Vals); source configurations remain separate from the canonical benchmark table.
Source rankSource model or configurationVals Terminal-Bench 2.1 mirror scoreSource setupParameters (B)Open / closed
1Claude Opus 5.5Anthropic87.6%Vals AI · Original resultNot reportedClosed
2GPT-6 AstraOpenAI87.3%max reasoning · Original resultNot reportedClosed
3GPT-5.6 SolOpenAI85.8%max reasoning · Original resultNot reportedClosed
4Claude Fable 5.1Anthropic85.0%Vals AI · Original resultNot reportedClosed
5Claude Opus 5Anthropic84.6%Vals AI · Original resultNot reportedClosed
6GPT-6 SolOpenAI83.1%max reasoning · Original resultNot reportedClosed
7Claude Sonnet 5.5Anthropic83.1%Vals AI · Original resultNot reportedClosed
8Gemini 3.8 FlashGoogle81.3%high reasoning · Original resultNot reportedClosed
9Kimi K3Moonshot AI80.9%Vals AI · Original resultNot reportedOpen
10Claude Fable 5Anthropic80.5%Vals AI · Original resultNot reportedClosed
11GPT-5.6 LunaOpenAI79.0%max reasoning · Original resultNot reportedClosed
12Muse Spark 1.3 MaxMeta79.0%max reasoning · Original resultNot reportedUnknown
13Grok 4.6xAI78.3%high reasoning · Original resultNot reportedClosed
14Gemini 3.7 FlashGoogle77.5%high reasoning · Original resultNot reportedClosed
15GPT-5.6 TerraOpenAI77.5%max reasoning · Original resultNot reportedClosed
16GPT-5.5OpenAI76.4%high reasoning · Original resultNot reportedClosed
17MiMo-V2.6-FlashXiaomi76.4%Vals AI · Original resultNot reportedOpen
18DeepSeek V4.1 Flash 1017DeepSeek74.5%high reasoning · Original resultNot reportedOpen
19Claude Sonnet 5Anthropic74.5%Vals AI · Original resultNot reportedClosed
20Gemini 3.5 FlashGoogle74.2%high reasoning · Original resultNot reportedClosed
21Gemini 3.6 FlashGoogle73.8%high reasoning · Original resultNot reportedClosed
22Grok 4.7xAI73.4%xhigh reasoning · Original resultNot reportedClosed
23GPT-6 LunaOpenAI73.0%max reasoning · Original resultNot reportedClosed
24Muse Spark 1.3Meta72.3%xhigh reasoning · Original resultNot reportedClosed
25Claude Opus 4.8Anthropic71.9%Vals AI · Original resultNot reportedClosed
26GLM-5.3Z.AI71.5%max reasoning · Original resultNot reportedOpen
27Gemini 3.1 Pro PreviewGoogle70.8%high reasoning · Original resultNot reportedUnknown
28Claude Opus 4.8 Claude CodeAnthropic69.7%Claude Code · Original resultNot reportedUnknown
29Muse Spark 1.2Meta69.7%xhigh reasoning · Original resultNot reportedClosed
30Muse Spark 1.1Meta69.3%xhigh reasoning · Original resultNot reportedClosed
31Claude Opus 4.7Anthropic68.5%Vals AI · Original resultNot reportedClosed
32Grok 4.5xAI67.8%high reasoning · Original resultNot reportedClosed
33GLM-5.2Z.AI67.8%max reasoning · Original resultNot reportedOpen
34MiMo-V2.6-ProXiaomi67.8%Vals AI · Original resultNot reportedOpen
35Qwen3.8 MaxAlibaba67.4%Vals AI · Original resultNot reportedOpen
36DeepSeek V4 Flash 0731DeepSeek67.0%Vals AI · Original resultNot reportedOpen
37Kimi K2.7 CodeMoonshot AI67.0%Vals AI · Original resultNot reportedOpen
38GLM-5.3-FlashZ.AI62.9%max reasoning · Original resultNot reportedOpen
39Qwen3.7 MaxAlibaba61.0%Vals AI · Original resultNot reportedClosed
40MiMo-V2.5Xiaomi60.7%Vals AI · Original resultNot reportedClosed
41Composer 2.5Cursor58.4%Cursor CLI · Original resultNot reportedClosed
42Qwen3.8-27BAlibaba58.4%xhigh reasoning · Original resultNot reportedOpen
43Gemini 4 ArgonGoogle57.7%high reasoning · Original resultNot reportedClosed
44GPT-5.5 CodexOpenAI57.3%Codex · Original resultNot reportedUnknown
45GPT-5.5 FactoryOpenAI57.3%Factory · Original resultNot reportedUnknown
46Claude Sonnet 4.6Anthropic57.3%Vals AI · Original resultNot reportedClosed
47MiMo-V2.5-ProXiaomi57.3%Vals AI · Original resultNot reportedClosed
48GLM-5.1Z.AI56.9%Vals AI · Original resultNot reportedOpen
49Inkling-SmallThinking Machines Lab55.1%0.99 reasoning · Original resultNot reportedOpen
50Hy4 previewTencent55.1%Vals AI · Original resultNot reportedOpen
51GPT-5.4 miniOpenAI54.7%high reasoning · Original resultNot reportedClosed
52DeepSeek V4 Pro 0813DeepSeek54.7%max reasoning · Original resultNot reportedUnknown
53Gemini 3 Flash PreviewGoogle53.9%high reasoning · Original resultNot reportedUnknown
54Kimi K2.6Moonshot AI53.6%Vals AI · Original resultNot reportedOpen
55MiniMax M3MiniMax53.6%Vals AI · Original resultNot reportedOpen
56Qwen3.6 PlusAlibaba53.2%Vals AI · Original resultNot reportedClosed
57Qwen3.7 PlusAlibaba52.8%Vals AI · Original resultNot reportedClosed
58Nemotron 3 Ultra 550b A55bNvidia50.9%Vals AI · Original resultNot reportedUnknown
59Gemini 3.5 Flash-LiteGoogle50.2%high reasoning · Original resultNot reportedClosed
60Ling 3.0 Flash 2607Ant50.2%Vals AI · Original resultNot reportedUnknown
61Ling 3.0 Flash Af Rc3Ant50.2%Vals AI · Original resultNot reportedUnknown
62DeepSeek V4 ProDeepSeek50.2%Vals AI · Original resultNot reportedOpen
63MiniMax M2.7MiniMax48.7%Vals AI · Original resultNot reportedOpen
64InklingThinking Machines Lab47.6%0.99 reasoning · Original resultNot reportedOpen
65Grok 4.20 0309 ReasoningxAI44.2%Vals AI · Original resultNot reportedUnknown
66Claude Haiku 4.5 ThinkingAnthropic43.8%Vals AI · Original resultNot reportedClosed
67Grok 4.3xAI41.9%high reasoning · Original resultNot reportedClosed
68Kimi K2.5 ThinkingMoonshot AI41.9%Vals AI · Original resultNot reportedUnknown
69GPT-5.4 nanoOpenAI41.6%high reasoning · Original resultNot reportedClosed
70Mistral Medium 3.5Mistral AI39.0%high reasoning · Original resultNot reportedUnknown
71Mercury 2.5Inception34.5%high reasoning · Original resultNot reportedClosed
72Gemini 3.1 Flash Lite PreviewGoogle34.1%high reasoning · Original resultNot reportedUnknown
73Laguna M.1Poolside34.1%Vals AI · Original resultNot reportedClosed
74Laguna XS.2Poolside25.8%Vals AI · Original resultNot reportedOpen
75Command A Plus 05 2026Cohere17.6%Vals AI · Original resultNot reportedUnknown
76Nemotron Lightning 3p5 30b A3bFireworks AI10.9%Vals AI · Original resultNot reportedUnknown

Older releaseReleased · 89 tasks · Official leaderboard · Not comparable with other releases

Terminal-Bench 2.0

Terminal-Bench 2.0 focuses on realistic CLI and repository workflows rather than toy code generation. BenchLM scores it from the agentic rows, which the benchmark page shows; this coding entry lists only the models whose coding record carries the result.

Terminal-Bench 2.0; source configurations remain separate from the canonical benchmark table.
Source rankSource model or configurationScoreWho ran itSource setupParameters (B)Open / closed
1GPT-5.5OpenAI82.0%Run by Terminal-Bench 2.0Terminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableGPT-5.3 CodexOpenAI77.3%Reported by OpenAITerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableGPT-5.4OpenAI75.1%Reported by OpenAITerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableQwen3.7 PlusAlibaba70.3%Reported by AlibabaTerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableQwen3.7 MaxAlibaba69.7%Reported by AlibabaTerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableClaude Opus 4.7Anthropic69.4%Reported by AnthropicTerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableComposer 2.5Cursor69.3%Reported by CursorTerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableMiMo-V2.5-ProXiaomi68.4%Reported by XiaomiTerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableDeepSeek V4 ProDeepSeek67.9%Reported by DeepSeekTerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableKimi K2.6Moonshot AI66.7%Reported by Moonshot AITerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableMiMo-V2.5Xiaomi65.8%Reported by XiaomiTerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableClaude Opus 4.6Anthropic65.4%Reported by AnthropicTerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableQwen 3.6 Max (preview)Alibaba65.4%Reported by AlibabaTerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableGLM-5.1Z.AI63.5%Reported by Z.AITerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableDeepSeek V4 Pro (High)DeepSeek63.3%Reported by DeepSeekTerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableComposer 2Cursor61.7%Reported by CursorTerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableQwen3.6 PlusAlibaba61.6%Reported by AlibabaTerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableGPT-5.4 miniOpenAI60.0%Reported by OpenAITerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableQwen3.6-27BAlibaba59.3%Reported by AlibabaTerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableClaude Opus 4.5Anthropic59.3%Reported by AnthropicTerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableClaude Sonnet 4.6Anthropic59.1%Reported by AnthropicTerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableMuse SparkMeta59.0%Reported by MetaTerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableMiniMax M2.7MiniMax57.0%Reported by MiniMaxTerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableDeepSeek V4 Flash 0731DeepSeek56.9%Reported by DeepSeekTerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableDeepSeek V4 Flash (High)DeepSeek56.6%Reported by DeepSeekTerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableGLM-5Z.AI56.2%Reported by Z.AITerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableHy3 PreviewTencent54.4%Reported by TencentTerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableQwen3.5 397BAlibaba52.5%Reported by AlibabaTerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableQwen3.6-35B-A3BAlibaba51.5%Reported by AlibabaTerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableKimi K2.5Moonshot AI50.8%Reported by Moonshot AITerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableClaude Sonnet 4.5Anthropic50.0%Reported by AnthropicTerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableQwen3.5-122B-A10BAlibaba49.4%Reported by AlibabaTerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableDeepSeek V4 FlashDeepSeek49.1%Reported by DeepSeekTerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableGrok 4.20xAI47.1%Reported by Meta AI (comparison table)Terminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableGPT-5.4 nanoOpenAI46.3%Reported by OpenAITerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableMAI-Thinking-1Microsoft46.0%Reported by MicrosoftTerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableLaguna M.1Poolside45.8%Reported by PoolsideTerminal-Bench contributors · Original resultNot reportedClosed
No rank on this tableQwen3.5-27BAlibaba41.6%Reported by AlibabaTerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableGLM-4.7Z.AI41.0%Reported by Z.AITerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableQwen3.5-35B-A3BAlibaba40.5%Reported by AlibabaTerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableLaguna XS 2.1Poolside37.5%Reported by PoolsideTerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableLaguna XS.2Poolside35.7%Reported by PoolsideTerminal-Bench contributors · Original resultNot reportedOpen
No rank on this tableLongCat-Flash-Lite-SparseMeituan33.7%Reported by MeituanTerminal-Bench contributors · Original resultNot reportedOpen

Vals Terminal-Bench 2.0 mirror

BenchLM mirrors this Vals AI leaderboard for reference only. It is excluded from BenchLM overall and category rankings.

Vals Terminal-Bench 2.0 mirror; source configurations remain separate from the canonical benchmark table.
Source rankSource model or configurationVals Terminal-Bench 2.0 mirror scoreSource setupParameters (B)Open / closed
1GPT-5.5OpenAI73.20%high reasoning · Original resultNot reportedClosed
2Claude Opus 4.8Anthropic70.04%Vals AI · Original resultNot reportedClosed
3Claude Opus 4.7Anthropic68.54%Vals AI · Original resultNot reportedClosed
4Gemini 3.5 FlashGoogle67.42%high reasoning · Original resultNot reportedClosed
5Gemini 3.1 Pro PreviewGoogle67.42%high reasoning · Original resultNot reportedUnknown
6GPT-5.3 CodexOpenAI64.05%high reasoning · Original resultNot reportedClosed
7Muse SparkMeta59.55%Vals AI · Original resultNot reportedClosed
8Claude Sonnet 4.6Anthropic59.55%Vals AI · Original resultNot reportedClosed
9Qwen3.7 MaxAlibaba59.18%Vals AI · Original resultNot reportedClosed
10Claude Opus 4.5Anthropic58.43%Vals AI · Original resultNot reportedClosed
11Claude Opus 4.6 (Adaptive)Anthropic58.43%Vals AI · Original resultNot reportedClosed
12GPT-5.4OpenAI58.43%high reasoning · Original resultNot reportedClosed
13Kimi K2.6Moonshot AI57.30%Vals AI · Original resultNot reportedOpen
14DeepSeek V4 ProDeepSeek56.18%Vals AI · Original resultNot reportedOpen
15Gemini 3 Pro PreviewGoogle55.06%high reasoning · Original resultNot reportedUnknown
16Claude Opus 4.5 ThinkingAnthropic53.93%Vals AI · Original resultNot reportedClosed
17GLM-5.1Z.AI53.93%Vals AI · Original resultNot reportedOpen
18Gemini 3 Flash PreviewGoogle51.69%high reasoning · Original resultNot reportedUnknown
19GPT-5.2OpenAI51.69%high reasoning · Original resultNot reportedClosed
20Qwen 3.6 Max (preview)Alibaba51.69%Vals AI · Original resultNot reportedClosed
21GLM 5 ThinkingZhipu AI49.44%Vals AI · Original resultNot reportedUnknown
22MiniMax M2.7MiniMax47.19%Vals AI · Original resultNot reportedOpen
23MiniMax M3MiniMax46.07%Vals AI · Original resultNot reportedOpen
24GPT-5.4 miniOpenAI44.94%high reasoning · Original resultNot reportedClosed
25Qwen3.6 PlusAlibaba44.94%Vals AI · Original resultNot reportedClosed
26GPT-5.1OpenAI44.94%high reasoning · Original resultNot reportedClosed
27Qwen3.6-27BAlibaba44.94%Vals AI · Original resultNot reportedOpen
28Grok 4.3xAI43.45%Vals AI · Original resultNot reportedClosed
29Claude Sonnet 4.5 ThinkingAnthropic41.57%Vals AI · Original resultNot reportedClosed
30MiniMax M2.5MiniMax41.57%Vals AI · Original resultNot reportedClosed
31Qwen3.5 Plus ThinkingAlibaba41.57%Vals AI · Original resultNot reportedUnknown
32Grok 4.20 0309 ReasoningxAI40.45%Vals AI · Original resultNot reportedUnknown
33Kimi K2.5 ThinkingMoonshot AI40.45%Vals AI · Original resultNot reportedUnknown
34GPT-5.4 nanoOpenAI39.89%high reasoning · Original resultNot reportedClosed
35Gemma 4 31b ItGoogle39.33%high reasoning · Original resultNot reportedUnknown
36Claude Haiku 4.5 ThinkingAnthropic38.20%Vals AI · Original resultNot reportedClosed
37GLM-4.7Z.AI38.20%Vals AI · Original resultNot reportedOpen
38MiniMax M2.1MiniMax37.08%Vals AI · Original resultNot reportedUnknown
39GPT-5OpenAI37.08%high reasoning · Original resultNot reportedClosed
40Kimi K2 ThinkingMoonshot AI37.08%Vals AI · Original resultNot reportedUnknown
41DeepSeek V3p2 ThinkingFireworks AI35.95%high reasoning · Original resultNot reportedUnknown
42DeepSeek V3p2Fireworks AI34.83%none reasoning · Original resultNot reportedUnknown
43Laguna M.1Poolside31.46%Vals AI · Original resultNot reportedClosed
44Gemini 2.5 ProGoogle30.34%Vals AI · Original resultNot reportedClosed
45Mistral Medium 3.5Mistral AI30.34%high reasoning · Original resultNot reportedUnknown
46Grok 4 Fast (Reasoning)xAI29.21%Vals AI · Original resultNot reportedClosed
47Laguna XS.2Poolside28.09%Vals AI · Original resultNot reportedOpen
48GLM-4.6Z.AI28.09%Vals AI · Original resultNot reportedOpen
49Grok 4 0709xAI28.09%Vals AI · Original resultNot reportedUnknown
50GPT-5 miniOpenAI26.97%high reasoning · Original resultNot reportedClosed
51Moonshotai Kimi K2 InstructTogether AI25.84%Vals AI · Original resultNot reportedUnknown
52Grok 4.1 Fast (Reasoning)xAI24.72%Vals AI · Original resultNot reportedClosed
53Gemini 3.1 Flash Lite PreviewGoogle24.72%high reasoning · Original resultNot reportedUnknown
54Qwen3.5 FlashAlibaba24.72%Vals AI · Original resultNot reportedClosed
55Qwen3 MaxAlibaba24.72%Vals AI · Original resultNot reportedClosed
56DeepSeek V3p1Fireworks AI22.47%Vals AI · Original resultNot reportedUnknown
57Gemini 2.5 Flash Preview 09 2025 ThinkingGoogle21.35%Vals AI · Original resultNot reportedUnknown
58Qwen3 MaxAlibaba20.23%Vals AI · Original resultNot reportedClosed
59GPT-OSS 120BOpenAI19.10%Vals AI · Original resultNot reportedOpen
60Grok 4.1 Fast Non ReasoningxAI17.98%Vals AI · Original resultNot reportedUnknown
61Trinity-Large-ThinkingArcee AI17.98%Vals AI · Original resultNot reportedOpen
62Mistral Small 2603Mistral AI16.85%high reasoning · Original resultNot reportedUnknown
63GPT-4.1OpenAI14.61%high reasoning · Original resultNot reportedClosed
64Magistral Medium 2509Mistral AI13.48%Vals AI · Original resultNot reportedUnknown
65Mistral Large 2512Mistral AI8.99%Vals AI · Original resultNot reportedUnknown
66Command A 03 2025Cohere2.25%Vals AI · Original resultNot reportedUnknown
67Llama4 Maverick Instruct BasicFireworks AI2.25%Vals AI · Original resultNot reportedUnknown

Older releaseReleased · 80 tasks at launch · Vals AI run · Not comparable with other releases

Vals Terminal-Bench 1.0 mirror

We keep the Vals-hosted Terminal-Bench 1.0 table separate from current Terminal-Bench 2.x and 3.0 results. It is historical, display-only context and does not enter weighted rankings.

Vals Terminal-Bench 1.0 mirror; source configurations remain separate from the canonical benchmark table.
Source rankSource model or configurationVals Terminal-Bench 1.0 mirror scoreSource setupParameters (B)Open / closed
1GPT-5.2OpenAI63.75%xhigh reasoning · Original resultNot reportedClosed
2Claude Sonnet 4.5 ThinkingAnthropic61.25%Vals AI · Original resultNot reportedClosed
3Gemini 3 Flash PreviewGoogle60.00%high reasoning · Original resultNot reportedUnknown
4GPT-5 CodexOpenAI58.75%high reasoning · Original resultNot reportedUnknown
5Claude Opus 4.5 ThinkingAnthropic57.50%Vals AI · Original resultNot reportedClosed
6GPT-5.1-CodexOpenAI57.50%high reasoning · Original resultNot reportedClosed
7Claude Opus 4.5Anthropic56.25%Vals AI · Original resultNot reportedClosed
8GPT-5.1-Codex-MaxOpenAI53.75%high reasoning · Original resultNot reportedClosed
9Gemini 3 Pro PreviewGoogle51.25%high reasoning · Original resultNot reportedUnknown
10Claude Haiku 4.5 ThinkingAnthropic50.00%Vals AI · Original resultNot reportedClosed
11DeepSeek V3p2Fireworks AI50.00%none reasoning · Original resultNot reportedUnknown
12GLM-4.7Z.AI50.00%Vals AI · Original resultNot reportedOpen
13GPT-5OpenAI48.75%high reasoning · Original resultNot reportedClosed
14GPT-5.1OpenAI47.50%high reasoning · Original resultNot reportedClosed
15Claude Sonnet 4 20250514 ThinkingAnthropic45.00%Vals AI · Original resultNot reportedUnknown
16Devstral 2512Mistral AI43.75%Vals AI · Original resultNot reportedUnknown
17GLM-4.6Z.AI42.50%Vals AI · Original resultNot reportedOpen
18GLM-4.5Z.AI41.25%Vals AI · Original resultNot reportedClosed
19Gemini 2.5 ProGoogle41.25%Vals AI · Original resultNot reportedClosed
20MiniMax M2.1MiniMax41.25%Vals AI · Original resultNot reportedUnknown
21DeepSeek V3p1Fireworks AI41.25%Vals AI · Original resultNot reportedUnknown
22Kimi K2 ThinkingMoonshot AI40.00%Vals AI · Original resultNot reportedUnknown
23Moonshotai Kimi K2 InstructTogether AI40.00%Vals AI · Original resultNot reportedUnknown
24Labs Devstral Small 2512Mistral AI40.00%Vals AI · Original resultNot reportedUnknown
25DeepSeek V3p2 ThinkingFireworks AI40.00%high reasoning · Original resultNot reportedUnknown
26Grok 4 0709xAI38.75%Vals AI · Original resultNot reportedUnknown
27Kimi K2 Instruct 0905Fireworks AI37.50%Vals AI · Original resultNot reportedUnknown
28Qwen3 Max PreviewAlibaba36.25%Vals AI · Original resultNot reportedUnknown
29Qwen3 MaxAlibaba36.25%Vals AI · Original resultNot reportedClosed
30GPT-4.1OpenAI33.75%high reasoning · Original resultNot reportedClosed
31Gemini 2.5 Flash Preview 09 2025Google31.25%Vals AI · Original resultNot reportedUnknown
32GPT-5 miniOpenAI30.00%high reasoning · Original resultNot reportedClosed
33Grok 4.1 Fast (Reasoning)xAI28.75%Vals AI · Original resultNot reportedClosed
34Magistral Medium 2509Mistral AI28.75%Vals AI · Original resultNot reportedUnknown
35Grok 4 Fast (Reasoning)xAI27.50%Vals AI · Original resultNot reportedClosed
36Gemini 2.5 Flash Preview 09 2025 ThinkingGoogle26.25%Vals AI · Original resultNot reportedUnknown
37GPT-OSS 120BOpenAI22.50%Vals AI · Original resultNot reportedOpen
38Grok 4.1 Fast Non ReasoningxAI21.25%Vals AI · Original resultNot reportedUnknown
39Mistral Large 2512Mistral AI21.25%Vals AI · Original resultNot reportedUnknown
40Gemini 2.5 Flash ThinkingGoogle21.25%Vals AI · Original resultNot reportedUnknown
41Grok Code Fast 1xAI20.00%Vals AI · Original resultNot reportedClosed
42Grok 4 Fast Non ReasoningxAI18.75%Vals AI · Original resultNot reportedUnknown
43Llama4 Maverick Instruct BasicFireworks AI17.50%Vals AI · Original resultNot reportedUnknown
44Magistral Small 2509Mistral AI15.00%Vals AI · Original resultNot reportedUnknown
45DeepSeek-R1DeepSeek13.75%Vals AI · Original resultNot reportedOpen
46Command A 03 2025Cohere6.25%Vals AI · Original resultNot reportedUnknown
47Jamba Large 1.7AI21 Labs6.25%Vals AI · Original resultNot reportedUnknown

How to read this leaderboard

Operator receipt: 63 sourced rows are currently displayable on this page; the leading published row is Copilot Fusion at 71.21%.

Honest limit: The table compares complete model-and-agent systems, not isolated base models. Errored trials, trials without stored artifacts, and 28 regrades attributed to verifier measurement noise keep their 4.0 scores, so 4.1 is a regrade of 4.0 rather than a fresh run.

How we use Terminal-Bench 4.1

Terminal-Bench 4.1 keeps the 66 tasks, 5 trials per task and 8-hour timeout of 4.0 and regrades the saved trials with hardened verifiers. Agent code now runs as an untrusted user in a separate process, and several over-strict checks accept valid alternative solutions. 162 of 11,550 trials changed score, and every entry stayed within its 4.0 95% confidence interval.

Each row keeps the harness and settings the source published. BenchLM shows the published table for reference; the per-model scores it stores feed the ranking.

Snapshot

63 system rows66 tasks5 trials per taskRegrade of 4.0Scored

The published Terminal-Bench 4.1 snapshot places Copilot Fusion first at 71.21%. The third row is 5.45 points behind. The broader top-10 range is 11.51 points, so the table still separates the published systems.

63 models have been evaluated on Terminal-Bench 4.1. The benchmark falls in the Agentic category. BenchLM shows the published table for reference; the per-model scores it stores feed the ranking. BenchAlign v5.8 gives Terminal-Bench 4.1 9% of the Agentic reference weight, so it moves the Agentic leaderboard and the overall ranking. Reference weights are relative weights in the calibrated model, not fixed shares of a score.

About Terminal-Bench 4.1

Year

2026

Tasks

66 professional computer-work tasks

Format

Task completion rate across 5 trials per task

Difficulty

Frontier autonomous knowledge work

Version 4.1 makes no agent-facing changes. It re-scores the saved 4.0 artifacts with updated verifiers: agent code now runs as an untrusted user in a separate process, and several over-strict checks accept valid alternative solutions. 162 of 11,550 trials changed score, and every one of the 35 leaderboard entries stayed within its 4.0 95% confidence interval. BenchLM keeps 4.1 on its own key so the 4.0 table remains a fixed record.

Freshness and provenance

Version

Terminal-Bench 4.1.0

Refresh cadence

Continuous

Staleness state

Current

Question availability

Public benchmark set

Current

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Questions

What does Terminal-Bench 4.1 measure?

The current Terminal-Bench release keeps the 66 professional computer-work tasks of 4.0 and regrades every saved trial with verifiers hardened against reward hacking and overly strict checks.

Which model leads the published Terminal-Bench 4.1 snapshot?

Copilot Fusion currently leads the published Terminal-Bench 4.1 snapshot with 71.21% tasks completed. BenchLM shows the published table for reference; the per-model scores it stores feed the ranking.

How many models are evaluated on Terminal-Bench 4.1?

The November 17, 2026 snapshot snapshot contains 63 AI models.

Last updated: November 17, 2026 snapshot · mirrored from the public benchmark leaderboard

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 7,000+ subscribers.

One email each week. Unsubscribe anytime.