Terminal-Bench 2.1
Data verified
Terminal-Bench 2.1 results, mostly provider-reported, stored separately from the Terminal-Bench 2.0 lane.
Current release
Terminal-Bench 4.0 is the current release. It changes task resources and the task set, so its freshly run scores are not directly comparable with this Terminal-Bench 2.x table.
View Terminal-Bench 4.0Top models on Terminal-Bench 2.1 — October 29, 2026
As of October 29, 2026, SWE-2 leads the Terminal-Bench 2.1 leaderboard with 92.8% , followed by GPT-5.6 Sol (91.9%) and DeepSeek V4.1 Flash 1017 (90.6%).
1Closed
SWE-2
Cognition
92.8%
Context 1M
2Closed
GPT-5.6 Sol
OpenAI
91.9%
Overall 74.83Context 1.05M
3Open weights
DeepSeek V4.1 Flash 1017
DeepSeek
90.6%
Overall 65.22Context 1M
82 modelsAgentic8% of Agentic reference weightCurrentUpdated October 29, 2026
Leaderboard (82 models)
82 of 82 rows
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | SWE-2Cognition | 92.8% | Not reported | Closed |
| 2 | GPT-5.6 SolOpenAI | 91.9% | Not reported | Closed |
| 3 | DeepSeek V4.1 Flash 1017DeepSeek | 90.6% | Not reported | Open |
| 4 | MiMo-V2.6-ProXiaomi | 89.9% | Not reported | Open |
| 5 | Gemini 3.8 FlashGoogle | 89.4% | Not reported | Closed |
| 6 | Muse Spark 1.3Meta | 88.8% | Not reported | Closed |
| 7 | Kimi K3Moonshot AI | 88.3% | Not reported | Open |
| 8 | GLM-5.3Z.AI | 88.2% | Not reported | Open |
| 9 | Claude Mythos 5Anthropic | 88.0% | Not reported | Closed |
| 10 | DeepSeek V4 Pro 0813DeepSeek | 87.9% | Not reported | Open |
| 11 | MiMo-V2.6-FlashXiaomi | 87.6% | Not reported | Open |
| 12 | GPT-5.6 TerraOpenAI | 87.4% | Not reported | Closed |
| 13 | Qwen3.8 MaxAlibaba | 86.6% | Not reported | Open |
| 14 | Ornith-1.5-397BOrnith AI | 86.1% | Not reported | Open |
| 15 | Gemini 3.7 FlashGoogle | 85.8% | Not reported | Closed |
| 16 | Hy4 previewTencent | 85.4% | Not reported | Open |
| 17 | Step 5 PreviewStepFun | 85.0% | Not reported | Open |
| 18 | GPT-5.6 LunaOpenAI | 84.7% | Not reported | Closed |
| 19 | Claude Fable 5Anthropic | 84.3% | Not reported | Closed |
| 20 | GLM-5.3-FlashZ.AI | 84.3% | Not reported | Open |
| 21 | Grok 4.5xAI | 83.3% | Not reported | Closed |
| 22 | Ultramarine 1 FlashPaper Instruments | 83.2% | Not reported | Closed |
| 23 | Muse Spark 1.2Meta | 82.9% | Not reported | Closed |
| 24 | DeepSeek V4 Flash 0731DeepSeek | 82.7% | Not reported | Open |
| 25 | Sakana Fugu-UltraSakana AI | 82.1% | Not reported | Closed |
| 26 | Ember-1Fireworks | 82.0% | Not reported | Closed |
| 27 | SWE-1.7Cognition | 81.5% | Not reported | Closed |
| 28 | Ling 3.1 FlashInclusionAI | 81.2% | Not reported | Open |
| 29 | GLM-5.2Z.AI | 81.0% | Not reported | Open |
| 30 | Claude Sonnet 5Anthropic | 80.4% | Not reported | Closed |
| 31 | Sakana FuguSakana AI | 80.2% | Not reported | Closed |
| 32 | Beam 1.0 501B-A23BReflection AI | 80.1% | Not reported | Open |
| 33 | Muse Spark 1.1Meta | 80.0% | Not reported | Closed |
| 34 | Particle-1Particle AI | 79.9% | Not reported | Closed |
| 35 | Beam 1.0 501B-A23B FP8Reflection AI | 78.9% | Not reported | Open |
| 36 | Beam 1.0 501B-A23B NVFP4Reflection AI | 78.4% | Not reported | Open |
| 37 | Atria Dawn PreviewShanghai Artificial Intelligence Laboratory | 78.3% | Not reported | Open |
| 38 | Ornith-1.0-397BDeepReinforce AI | 77.5% | Not reported | Open |
| 39 | Gemini 3.5 FlashGoogle | 76.2% | Not reported | Closed |
| 40 | dots3-note PreviewDots Studio | 75.1% | Not reported | Open |
| 41 | Claude Opus 4.8Anthropic | 74.6% | Not reported | Closed |
| 42 | Jais-3-375B-A23BIFM | 73.8% | Not reported | Open |
| 43 | Qwen3.8-27BAlibaba | 73.0% | Not reported | Open |
| 44 | OrcaCyber-Nano-1.0OrcaRouter | 72.5% | Not reported | Open |
| 45 | Seed 2.1 ProByteDance | 71.0% | Not reported | Closed |
| 46 | Apodex 1.1Apodex | 70.8% | Not reported | Closed |
| 47 | Laguna S 2.1Poolside | 70.2% | Not reported | Open |
| 48 | Quasar 438BMultiverse Computing | 69.3% | Not reported | Closed |
| 49 | Ornith-1.5-35B-A3BOrnith AI | 67.8% | Not reported | Open |
| 50 | Seed 2.1 TurboByteDance | 67.6% | Not reported | Closed |
| 51 | Jais-3-MoVA-36B-A4BIFM | 67.3% | Not reported | Open |
| 52 | Nemotron 3.5 Super VL 120B A12BNVIDIA | 66.7% | Not reported | Open |
| 53 | MiniMax M3MiniMax | 66.0% | Not reported | Open |
| 54 | Pokee-Isaac 28BPokee AI | 65.1% | Not reported | Closed |
| 55 | Inkling-SmallThinking Machines Lab | 64.7% | Not reported | Open |
| 56 | Ornith-1.0-35BDeepReinforce AI | 64.2% | Not reported | Open |
| 57 | InklingThinking Machines Lab | 63.8% | Not reported | Open |
| 58 | Jais-3-32BIFM | 63.7% | Not reported | Open |
| 59 | MAI-Code-1.1-FlashMicrosoft | 62.9% | Not reported | Closed |
| 60 | Nemotron 3.5 Super VL 120B A12B NVFP4NVIDIA | 60.8% | Not reported | Open |
| 61 | Step 3.7 FlashStepFun | 59.5% | Not reported | Open |
| 62 | Ling 3.0 FlashInclusionAI | 57.0% | Not reported | Open |
| 63 | Solar Pro 4Upstage | 57.0% | Not reported | Closed |
| 64 | Nemotron 3 UltraNVIDIA | 56.4% | Not reported | Open |
| 65 | FlintRecursive | 54.9% | Not reported | Open |
| 66 | Gemini 3.5 Flash-LiteGoogle | 54.0% | Not reported | Closed |
| 67 | Nemotron 3.5 Bolt 74B-A9B BF16NVIDIA | 53.9% | Not reported | Open |
| 68 | Ternary Bonsai 2 27BPrism ML | 52.8% | Not reported | Open |
| 69 | Nemotron 3.5 Bolt 74B-A9BNVIDIA | 51.0% | Not reported | Open |
| 70 | Jais-3-7BIFM | 46.4% | Not reported | Open |
| 71 | Ornith-1.5-9BOrnith AI | 46.2% | Not reported | Open |
| 72 | K-EXAONE 2.0LG AI Research | 43.8% | Not reported | Open |
| 73 | Ornith-1.0-9BDeepReinforce AI | 43.1% | Not reported | Open |
| 74 | Jais-3-3.7BIFM | 41.2% | Not reported | Open |
| 75 | A.X K2SK Telecom | 36.0% | Not reported | Open |
| 76 | Granite 4.2 30BIBM | 29.2% | Not reported | Open |
| 77 | Mercury 2Inception | 27.0% | Not reported | Closed |
| 78 | Nemotron 3.5 Lightning 30B A3B NVFP4NVIDIA | 23.5% | Not reported | Open |
| 79 | Granite 4.2 8BIBM | 20.6% | Not reported | Open |
| 80 | Mellum2.1-12B-A2.5B-ThinkingJetBrains | 17.4% | Not reported | Open |
| 81 | LFM3-1.7BLiquidAI | 12.9% | Not reported | Open |
| 82 | MiniCPM5-2BOpenBMB | 8.6% | Not reported | Open |
AA Terminal-Bench 2.1
Stored separately from the core Terminal-Bench lane because Artificial Analysis controls the evaluation configuration.
21 of 21 rows
| Source rank | Source model or configuration | Score | Source setup | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | Claude Fable 5.1Anthropic | 91.4% | Artificial AnalysisOriginal result | Not reported | Closed |
| 2 | Claude Opus 5Anthropic | 89.1% | Artificial AnalysisOriginal result | Not reported | Closed |
| 3 | GPT-6 AstraOpenAI | 88.4% | Artificial AnalysisOriginal result | Not reported | Closed |
| 4 | Grok 4.6xAI | 88.4% | Artificial AnalysisOriginal result | Not reported | Closed |
| 5 | GPT-5.6 TerraOpenAI | 88.0% | Artificial AnalysisOriginal result | Not reported | Closed |
| 6 | Gemini 3.8 FlashGoogle | 87.6% | Artificial AnalysisOriginal result | Not reported | Closed |
| 7 | Qwen3.8-Flash-NextAlibaba | 86.1% | Artificial AnalysisOriginal result | Not reported | Open |
| 8 | Gemini 3.7 FlashGoogle | 85.8% | Artificial AnalysisOriginal result | Not reported | Closed |
| 9 | Kimi K3Moonshot AI | 85.0% | Artificial AnalysisOriginal result | Not reported | Open |
| 10 | Claude Fable 5Anthropic | 84.6% | Artificial AnalysisOriginal result | Not reported | Closed |
| 11 | Claude Opus 4.8Anthropic | 84.6% | Artificial AnalysisOriginal result | Not reported | Closed |
| 12 | GPT-5.5OpenAI | 84.3% | Artificial AnalysisOriginal result | Not reported | Closed |
| 13 | GLM-5.3-FlashZ.AI | 84.3% | Artificial AnalysisOriginal result | Not reported | Open |
| 14 | GLM-5.3Z.AI | 83.9% | Artificial AnalysisOriginal result | Not reported | Open |
| 15 | Claude Opus 4.7 (Adaptive)Anthropic | 83.1% | Artificial AnalysisOriginal result | Not reported | Closed |
| 16 | Grok 4.5xAI | 81.6% | Artificial AnalysisOriginal result | Not reported | Closed |
| 17 | GPT-5.6 LunaOpenAI | 80.9% | Artificial AnalysisOriginal result | Not reported | Closed |
| 18 | Claude Sonnet 5Anthropic | 80.5% | Artificial AnalysisOriginal result | Not reported | Closed |
| 19 | Qwen3.8-27BAlibaba | 79.8% | Artificial AnalysisOriginal result | Not reported | Open |
| 20 | Muse Spark OpenMeta | 76.4% | Artificial AnalysisOriginal result | Not reported | Open |
| 21 | InklingThinking Machines Lab | 55.1% | Artificial AnalysisOriginal result | Not reported | Open |
Terminal-Bench 2.1 (Vals)
BenchLM mirrors this Vals AI leaderboard for reference only. It is excluded from BenchLM overall and category rankings.
76 of 76 rows
| Source rank | Source model or configuration | Vals Terminal-Bench 2.1 mirror score | Source setup | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5Anthropic | 87.6% | Vals AIOriginal result | Not reported | Closed |
| 2 | GPT-6 AstraOpenAI | 87.3% | max reasoningOriginal result | Not reported | Closed |
| 3 | GPT-5.6 SolOpenAI | 85.8% | max reasoningOriginal result | Not reported | Closed |
| 4 | Claude Fable 5.1Anthropic | 85.0% | Vals AIOriginal result | Not reported | Closed |
| 5 | Claude Opus 5Anthropic | 84.6% | Vals AIOriginal result | Not reported | Closed |
| 6 | GPT-6 SolOpenAI | 83.1% | max reasoningOriginal result | Not reported | Closed |
| 7 | Claude Sonnet 5.5Anthropic | 83.1% | Vals AIOriginal result | Not reported | Closed |
| 8 | Gemini 3.8 FlashGoogle | 81.3% | high reasoningOriginal result | Not reported | Closed |
| 9 | Kimi K3Moonshot AI | 80.9% | Vals AIOriginal result | Not reported | Open |
| 10 | Claude Fable 5Anthropic | 80.5% | Vals AIOriginal result | Not reported | Closed |
| 11 | GPT-5.6 LunaOpenAI | 79.0% | max reasoningOriginal result | Not reported | Closed |
| 12 | Muse Spark 1.3 MaxMeta | 79.0% | max reasoningOriginal result | Not reported | Unknown |
| 13 | Grok 4.6xAI | 78.3% | high reasoningOriginal result | Not reported | Closed |
| 14 | Gemini 3.7 FlashGoogle | 77.5% | high reasoningOriginal result | Not reported | Closed |
| 15 | GPT-5.6 TerraOpenAI | 77.5% | max reasoningOriginal result | Not reported | Closed |
| 16 | GPT-5.5OpenAI | 76.4% | high reasoningOriginal result | Not reported | Closed |
| 17 | MiMo-V2.6-FlashXiaomi | 76.4% | Vals AIOriginal result | Not reported | Open |
| 18 | DeepSeek V4.1 Flash 1017DeepSeek | 74.5% | high reasoningOriginal result | Not reported | Open |
| 19 | Claude Sonnet 5Anthropic | 74.5% | Vals AIOriginal result | Not reported | Closed |
| 20 | Gemini 3.5 FlashGoogle | 74.2% | high reasoningOriginal result | Not reported | Closed |
| 21 | Gemini 3.6 FlashGoogle | 73.8% | high reasoningOriginal result | Not reported | Closed |
| 22 | Grok 4.7xAI | 73.4% | xhigh reasoningOriginal result | Not reported | Closed |
| 23 | GPT-6 LunaOpenAI | 73.0% | max reasoningOriginal result | Not reported | Closed |
| 24 | Muse Spark 1.3Meta | 72.3% | xhigh reasoningOriginal result | Not reported | Closed |
| 25 | Claude Opus 4.8Anthropic | 71.9% | Vals AIOriginal result | Not reported | Closed |
| 26 | GLM-5.3Z.AI | 71.5% | max reasoningOriginal result | Not reported | Open |
| 27 | Gemini 3.1 Pro PreviewGoogle | 70.8% | high reasoningOriginal result | Not reported | Unknown |
| 28 | Claude Opus 4.8 Claude CodeAnthropic | 69.7% | Claude CodeOriginal result | Not reported | Unknown |
| 29 | Muse Spark 1.2Meta | 69.7% | xhigh reasoningOriginal result | Not reported | Closed |
| 30 | Muse Spark 1.1Meta | 69.3% | xhigh reasoningOriginal result | Not reported | Closed |
| 31 | Claude Opus 4.7Anthropic | 68.5% | Vals AIOriginal result | Not reported | Closed |
| 32 | Grok 4.5xAI | 67.8% | high reasoningOriginal result | Not reported | Closed |
| 33 | GLM-5.2Z.AI | 67.8% | max reasoningOriginal result | Not reported | Open |
| 34 | MiMo-V2.6-ProXiaomi | 67.8% | Vals AIOriginal result | Not reported | Open |
| 35 | Qwen3.8 MaxAlibaba | 67.4% | Vals AIOriginal result | Not reported | Open |
| 36 | DeepSeek V4 Flash 0731DeepSeek | 67.0% | Vals AIOriginal result | Not reported | Open |
| 37 | Kimi K2.7 CodeMoonshot AI | 67.0% | Vals AIOriginal result | Not reported | Open |
| 38 | GLM-5.3-FlashZ.AI | 62.9% | max reasoningOriginal result | Not reported | Open |
| 39 | Qwen3.7 MaxAlibaba | 61.0% | Vals AIOriginal result | Not reported | Closed |
| 40 | MiMo-V2.5Xiaomi | 60.7% | Vals AIOriginal result | Not reported | Closed |
| 41 | Composer 2.5Cursor | 58.4% | Cursor CLIOriginal result | Not reported | Closed |
| 42 | Qwen3.8-27BAlibaba | 58.4% | xhigh reasoningOriginal result | Not reported | Open |
| 43 | Gemini 4 ArgonGoogle | 57.7% | high reasoningOriginal result | Not reported | Closed |
| 44 | GPT-5.5 CodexOpenAI | 57.3% | CodexOriginal result | Not reported | Unknown |
| 45 | GPT-5.5 FactoryOpenAI | 57.3% | FactoryOriginal result | Not reported | Unknown |
| 46 | Claude Sonnet 4.6Anthropic | 57.3% | Vals AIOriginal result | Not reported | Closed |
| 47 | MiMo-V2.5-ProXiaomi | 57.3% | Vals AIOriginal result | Not reported | Closed |
| 48 | GLM-5.1Z.AI | 56.9% | Vals AIOriginal result | Not reported | Open |
| 49 | Inkling-SmallThinking Machines Lab | 55.1% | 0.99 reasoningOriginal result | Not reported | Open |
| 50 | Hy4 previewTencent | 55.1% | Vals AIOriginal result | Not reported | Open |
| 51 | GPT-5.4 miniOpenAI | 54.7% | high reasoningOriginal result | Not reported | Closed |
| 52 | DeepSeek V4 Pro 0813DeepSeek | 54.7% | max reasoningOriginal result | Not reported | Open |
| 53 | Gemini 3 Flash PreviewGoogle | 53.9% | high reasoningOriginal result | Not reported | Unknown |
| 54 | Kimi K2.6Moonshot AI | 53.6% | Vals AIOriginal result | Not reported | Open |
| 55 | MiniMax M3MiniMax | 53.6% | Vals AIOriginal result | Not reported | Open |
| 56 | Qwen3.6 PlusAlibaba | 53.2% | Vals AIOriginal result | Not reported | Closed |
| 57 | Qwen3.7 PlusAlibaba | 52.8% | Vals AIOriginal result | Not reported | Closed |
| 58 | Nemotron 3 Ultra 550b A55bNvidia | 50.9% | Vals AIOriginal result | Not reported | Unknown |
| 59 | Gemini 3.5 Flash-LiteGoogle | 50.2% | high reasoningOriginal result | Not reported | Closed |
| 60 | Ling 3.0 Flash 2607Ant | 50.2% | Vals AIOriginal result | Not reported | Unknown |
| 61 | Ling 3.0 Flash Af Rc3Ant | 50.2% | Vals AIOriginal result | Not reported | Unknown |
| 62 | DeepSeek V4 Pro 0813DeepSeek | 50.2% | Vals AIOriginal result | Not reported | Open |
| 63 | MiniMax M2.7MiniMax | 48.7% | Vals AIOriginal result | Not reported | Open |
| 64 | InklingThinking Machines Lab | 47.6% | 0.99 reasoningOriginal result | Not reported | Open |
| 65 | Grok 4.20 0309 ReasoningxAI | 44.2% | Vals AIOriginal result | Not reported | Unknown |
| 66 | Claude Haiku 4.5 ThinkingAnthropic | 43.8% | Vals AIOriginal result | Not reported | Closed |
| 67 | Grok 4.3xAI | 41.9% | high reasoningOriginal result | Not reported | Closed |
| 68 | Kimi K2.5 ThinkingMoonshot AI | 41.9% | Vals AIOriginal result | Not reported | Unknown |
| 69 | GPT-5.4 nanoOpenAI | 41.6% | high reasoningOriginal result | Not reported | Closed |
| 70 | Mistral Medium 3.5Mistral AI | 39.0% | high reasoningOriginal result | Not reported | Unknown |
| 71 | Mercury 2.5Inception | 34.5% | high reasoningOriginal result | Not reported | Closed |
| 72 | Gemini 3.1 Flash Lite PreviewGoogle | 34.1% | high reasoningOriginal result | Not reported | Unknown |
| 73 | Laguna M.1Poolside | 34.1% | Vals AIOriginal result | Not reported | Closed |
| 74 | Laguna XS.2Poolside | 25.8% | Vals AIOriginal result | Not reported | Open |
| 75 | Command A Plus 05 2026Cohere | 17.6% | Vals AIOriginal result | Not reported | Unknown |
| 76 | Nemotron Lightning 3p5 30b A3bFireworks AI | 10.9% | Vals AIOriginal result | Not reported | Unknown |
According to BenchLM.ai, SWE-2 leads the Terminal-Bench 2.1 benchmark with a score of 92.8%, followed by GPT-5.6 Sol (91.9%) and DeepSeek V4.1 Flash 1017 (90.6%). The top models are clustered within 2.2 points, suggesting this benchmark is nearing saturation for frontier models.
82 models have been evaluated on Terminal-Bench 2.1. The benchmark falls in the Agentic category. BenchAlign v5.8 gives Terminal-Bench 2.1 8% 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 2.1
Year
2026
Tasks
Terminal-based software-agent tasks
Format
Interactive task success rate
Difficulty
Professional software engineering
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.
Freshness and provenance
Version
Terminal-Bench 2.1 2026
Refresh cadence
Quarterly
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 2.1 measure?
Terminal-Bench 2.1 results, mostly provider-reported, stored separately from the Terminal-Bench 2.0 lane.
Which model scores highest on Terminal-Bench 2.1?
SWE-2 by Cognition currently leads with a score of 92.8% on Terminal-Bench 2.1.
How many models are evaluated on Terminal-Bench 2.1?
82 AI models have published results on Terminal-Bench 2.1 in the BenchLM catalog.
Compare top models on Terminal-Bench 2.1
Last updated: October 29, 2026 · BenchLM version Terminal-Bench 2.1 2026
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.