Skip to main content

Vals Terminal-Bench 2.1 (Terminal-Bench 2.1)

State-of-the-art set of difficult terminal-based tasks

Data verified

How BenchLM shows Vals Terminal-Bench 2.1 mirror

BenchLM mirrors the public Vals AI Vals Terminal-Bench 2.1 mirror leaderboard captured from https://www.vals.ai/benchmarks/terminal-bench-2-1 and updated by Vals on July 9, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.

Vals Terminal-Bench 2.1 mirror is display only on BenchLM. Vals proprietary or Vals-hosted aggregate views are useful context, but BenchLM does not use them as weighted ranking inputs or as a replacement for benchmark-native source records.

41 Vals rows4 task viewspublic datasetTasks: Overall, Easy, Medium, HardDisplay only

Vals Terminal-Bench 2.1 mirror score on Terminal-Bench 2.1 — July 9, 2026

BenchLM mirrors the published vals terminal-bench 2.1 mirror score view for Terminal-Bench 2.1. GPT-5.6 Sol leads the public snapshot at 85.77% , followed by Claude Fable 5 (80.52%) and GPT-5.6 Luna (79.03%). BenchLM does not use these results to rank models overall.

41 modelsExternal benchmark mirrorsCurrentDisplay onlyUpdated July 9, 2026

The published Terminal-Bench 2.1 snapshot is tightly clustered at the top: GPT-5.6 Sol sits at 85.77%, while the third row is only 6.74 points behind. The broader top-10 spread is 16.11 points, so the benchmark still separates strong models even when the leaders cluster.

41 models have been evaluated on Terminal-Bench 2.1. The benchmark falls in the External benchmark mirrors category. BenchLM tracks this category separately from its weighted global scoring system, so these results are best compared on the dedicated Korean benchmark views. Terminal-Bench 2.1 is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About Terminal-Bench 2.1

Year

2026

Tasks

Terminal-based task execution

Format

Accuracy score

Difficulty

Frontier terminal-agent execution

BenchLM mirrors the public Vals AI Terminal-Bench 2.1 leaderboard as display-only external evidence. The captured snapshot preserves overall scores, task-level scores where Vals publishes them, uncertainty, latency, and cost-per-test metadata. It is excluded from BenchLM weighted rankings.

BenchLM freshness & provenance

Version

Terminal-Bench 2.1 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

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.

Vals Terminal-Bench 2.1 mirror score table (41 models)

1
GPT-5.6 Solopenai/gpt-5.6-sol
85.77%
2
Claude Fable 5anthropic/claude-fable-5
80.52%
3
GPT-5.6 Lunaopenai/gpt-5.6-luna
79.03%
4
GPT-5.5openai/gpt-5.5
76.40%
5
Claude Sonnet 5anthropic/claude-sonnet-5
74.53%
6
Gemini 3.5 Flashgoogle/gemini-3.5-flash
74.16%
7
GPT-5.6 Terraopenai/gpt-5.6-terra
73.41%
8
Claude Opus 4.8anthropic/claude-opus-4-8
71.91%
9
Gemini 3.1 Pro Previewgoogle/gemini-3.1-pro-preview
70.79%
10
Claude Opus 4.8 Claude Codeanthropic/claude-opus-4-8-claude-code
69.66%
11
Muse Spark 1 1meta/muse_spark_1_1
69.29%
12
Claude Opus 4.7anthropic/claude-opus-4-7
68.54%
13
Grok 4.5grok/grok-4.5
67.79%
14
GLM 5.2zai/glm-5.2
67.79%
15
Kimi K2.7 Codekimi/kimi-k2.7-code
67.04%
16
Qwen3.7 Maxalibaba/qwen3.7-max
61.05%
17
Mimo V2.5xiaomi/mimo-v2.5
60.67%
18
Composer 2.5cursor/composer-2.5
58.43%
19
GPT-5.5 Codexopenai/gpt-5.5-codex
57.30%
20
GPT-5.5 Factoryopenai/gpt-5.5-factory
57.30%
21
Claude Sonnet 4.6anthropic/claude-sonnet-4-6
57.30%
22
Mimo V2.5 Proxiaomi/mimo-v2.5-pro
57.30%
23
GLM 5.1zai/glm-5.1
56.93%
24
GPT-5.4 Miniopenai/gpt-5.4-mini-2026-03-17
54.68%
25
Gemini 3 Flash Previewgoogle/gemini-3-flash-preview
53.93%
26
Kimi K2.6kimi/kimi-k2.6
53.56%
27
MiniMax M3minimax/MiniMax-M3
53.56%
28
Qwen3.6 Plusalibaba/qwen3.6-plus
53.18%
29
Qwen3.7 Plusalibaba/qwen3.7-plus
52.81%
30
Nemotron 3 Ultra 550b A55bnvidia/nemotron-3-ultra-550b-a55b
50.94%
31
DeepSeek V4 Prodeepseek/deepseek-v4-pro
50.19%
32
MiniMax M2.7minimax/MiniMax-M2.7
48.69%
33
Grok 4.20 0309 Reasoninggrok/grok-4.20-0309-reasoning
44.20%
34
Claude Haiku 4.5 20251001 Thinkinganthropic/claude-haiku-4-5-20251001-thinking
43.82%
35
Grok 4.3grok/grok-4.3
41.95%
36
GPT-5.4 Nanoopenai/gpt-5.4-nano-2026-03-17
41.57%
37
Mistral Medium 3.5mistralai/mistral-medium-3.5
38.95%
38
Gemini 3.1 Flash Lite Previewgoogle/gemini-3.1-flash-lite-preview
34.08%
39
Laguna M.1poolside/laguna-m.1
31.84%
40
Laguna Xs.2poolside/laguna-xs.2
26.97%
41
Command A Plus 05 2026cohere/command-a-plus-05-2026
17.60%

FAQ

What does Terminal-Bench 2.1 measure?

State-of-the-art set of difficult terminal-based tasks

Which model leads the published Terminal-Bench 2.1 snapshot?

GPT-5.6 Sol currently leads the published Terminal-Bench 2.1 snapshot with 85.77% vals terminal-bench 2.1 mirror score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on Terminal-Bench 2.1?

41 AI models are included in BenchLM's mirrored Terminal-Bench 2.1 snapshot, based on the public leaderboard captured on July 9, 2026.

Last updated: July 9, 2026 · mirrored from the public benchmark leaderboard

The AI models change fast. We track them for you.

For engineers, researchers, and the plain curious — a weekly brief on new models, ranking shifts, and pricing changes.

Free. No spam. Unsubscribe anytime.