Terminal-Bench 2.0
A benchmark for agentic software engineering tasks executed in real terminal environments. Models must inspect files, run commands, edit code, and recover from errors over multi-step workflows.
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 2.0 — October 29, 2026
As of October 29, 2026, GPT-5.5 leads the Terminal-Bench 2.0 leaderboard with 82% , followed by GPT-5.3 Codex (77.3%) and GPT-5.4 (75.1%).
GPT-5.5
OpenAI
GPT-5.3 Codex
OpenAI
GPT-5.4
OpenAI
45 modelsAgentic7% of Agentic reference weightCurrentUpdated October 29, 2026
Leaderboard (45 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | GPT-5.5OpenAI | 82% | Not reported | Closed |
| 2 | GPT-5.3 CodexOpenAI | 77.3% | Not reported | Closed |
| 3 | GPT-5.4OpenAI | 75.1% | Not reported | Closed |
| 4 | Qwen3.7 PlusAlibaba | 70.3% | Not reported | Closed |
| 5 | Qwen3.7 MaxAlibaba | 69.7% | Not reported | Closed |
| 6 | Claude Opus 4.7 (Adaptive)Anthropic | 69.4% | Not reported | Closed |
| 7 | Composer 2.5Cursor | 69.3% | Not reported | Closed |
| 8 | MiMo-V2.5-ProXiaomi | 68.4% | Not reported | Closed |
| 9 | DeepSeek V4 Pro 0813DeepSeek | 67.9% | Not reported | Open |
| 10 | Kimi K2.6Moonshot AI | 66.7% | Not reported | Open |
| 11 | MiMo-V2.5Xiaomi | 65.8% | Not reported | Closed |
| 12 | Claude Opus 4.6Anthropic | 65.4% | Not reported | Closed |
| 13 | Qwen 3.6 Max (preview)Alibaba | 65.4% | Not reported | Closed |
| 14 | GLM-5.1Z.AI | 63.5% | Not reported | Open |
| 15 | DeepSeek V4 Pro (High)DeepSeek | 63.3% | Not reported | Open |
| 16 | Composer 2Cursor | 61.7% | Not reported | Closed |
| 17 | Qwen3.6 PlusAlibaba | 61.6% | Not reported | Closed |
| 18 | GPT-5.4 miniOpenAI | 60% | Not reported | Closed |
| 19 | Qwen3.6-27BAlibaba | 59.3% | Not reported | Open |
| 20 | Claude Opus 4.5Anthropic | 59.3% | Not reported | Closed |
| 21 | Claude Sonnet 4.6Anthropic | 59.1% | Not reported | Closed |
| 22 | DeepSeek V4 ProDeepSeek | 59.1% | Not reported | Open |
| 23 | Muse SparkMeta | 59% | Not reported | Closed |
| 24 | MiniMax M2.7MiniMax | 57% | Not reported | Open |
| 25 | DeepSeek V4 Flash 0731DeepSeek | 56.9% | Not reported | Open |
| 26 | DeepSeek V4 Flash (High)DeepSeek | 56.6% | Not reported | Open |
| 27 | GLM-5Z.AI | 56.2% | Not reported | Open |
| 28 | Hy3 PreviewTencent | 54.4% | Not reported | Open |
| 29 | Qwen3.5 397BAlibaba | 52.5% | Not reported | Open |
| 30 | Qwen3.6-35B-A3BAlibaba | 51.5% | Not reported | Open |
| 31 | Kimi K2.5 (Reasoning)Moonshot AI | 50.8% | Not reported | Closed |
| 32 | Kimi K2.5Moonshot AI | 50.8% | Not reported | Open |
| 33 | Claude Sonnet 4.5Anthropic | 50% | Not reported | Closed |
| 34 | Qwen3.5-122B-A10BAlibaba | 49.4% | Not reported | Open |
| 35 | DeepSeek V4 FlashDeepSeek | 49.1% | Not reported | Open |
| 36 | Grok 4.20xAI | 47.1% | Not reported | Closed |
| 37 | GPT-5.4 nanoOpenAI | 46.3% | Not reported | Closed |
| 38 | MAI-Thinking-1Microsoft | 46% | Not reported | Closed |
| 39 | Laguna M.1Poolside | 45.8% | Not reported | Closed |
| 40 | Qwen3.5-27BAlibaba | 41.6% | Not reported | Open |
| 41 | GLM-4.7Z.AI | 41% | Not reported | Open |
| 42 | Qwen3.5-35B-A3BAlibaba | 40.5% | Not reported | Open |
| 43 | Laguna XS 2.1Poolside | 37.5% | Not reported | Open |
| 44 | Laguna XS.2Poolside | 35.7% | Not reported | Open |
| 45 | LongCat-Flash-Lite-SparseMeituan | 33.7% | Not reported | Open |
Vals Terminal-Bench 2.0 mirror
BenchLM mirrors this Vals AI leaderboard for reference only. It is excluded from BenchLM overall and category rankings.
| Source rank | Source model or configuration | Vals Terminal-Bench 2.0 mirror score | Source setup | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | GPT-5.5OpenAI | 73.20% | high reasoningOriginal result | Not reported | Closed |
| 2 | Claude Opus 4.8Anthropic | 70.04% | Vals AIOriginal result | Not reported | Closed |
| 3 | Claude Opus 4.7Anthropic | 68.54% | Vals AIOriginal result | Not reported | Closed |
| 4 | Gemini 3.5 FlashGoogle | 67.42% | high reasoningOriginal result | Not reported | Closed |
| 5 | Gemini 3.1 Pro PreviewGoogle | 67.42% | high reasoningOriginal result | Not reported | Unknown |
| 6 | GPT-5.3 CodexOpenAI | 64.05% | high reasoningOriginal result | Not reported | Closed |
| 7 | Muse SparkMeta | 59.55% | Vals AIOriginal result | Not reported | Closed |
| 8 | Claude Sonnet 4.6Anthropic | 59.55% | Vals AIOriginal result | Not reported | Closed |
| 9 | Qwen3.7 MaxAlibaba | 59.18% | Vals AIOriginal result | Not reported | Closed |
| 10 | Claude Opus 4.5Anthropic | 58.43% | Vals AIOriginal result | Not reported | Closed |
| 11 | Claude Opus 4.6 (Adaptive)Anthropic | 58.43% | Vals AIOriginal result | Not reported | Closed |
| 12 | GPT-5.4OpenAI | 58.43% | high reasoningOriginal result | Not reported | Closed |
| 13 | Kimi K2.6Moonshot AI | 57.30% | Vals AIOriginal result | Not reported | Open |
| 14 | DeepSeek V4 Pro 0813DeepSeek | 56.18% | Vals AIOriginal result | Not reported | Open |
| 15 | Gemini 3 Pro PreviewGoogle | 55.06% | high reasoningOriginal result | Not reported | Unknown |
| 16 | Claude Opus 4.5 ThinkingAnthropic | 53.93% | Vals AIOriginal result | Not reported | Closed |
| 17 | GLM-5.1Z.AI | 53.93% | Vals AIOriginal result | Not reported | Open |
| 18 | Gemini 3 Flash PreviewGoogle | 51.69% | high reasoningOriginal result | Not reported | Unknown |
| 19 | GPT-5.2OpenAI | 51.69% | high reasoningOriginal result | Not reported | Closed |
| 20 | Qwen 3.6 Max (preview)Alibaba | 51.69% | Vals AIOriginal result | Not reported | Closed |
| 21 | GLM 5 ThinkingZhipu AI | 49.44% | Vals AIOriginal result | Not reported | Unknown |
| 22 | MiniMax M2.7MiniMax | 47.19% | Vals AIOriginal result | Not reported | Open |
| 23 | MiniMax M3MiniMax | 46.07% | Vals AIOriginal result | Not reported | Open |
| 24 | GPT-5.4 miniOpenAI | 44.94% | high reasoningOriginal result | Not reported | Closed |
| 25 | Qwen3.6 PlusAlibaba | 44.94% | Vals AIOriginal result | Not reported | Closed |
| 26 | GPT-5.1OpenAI | 44.94% | high reasoningOriginal result | Not reported | Closed |
| 27 | Qwen3.6-27BAlibaba | 44.94% | Vals AIOriginal result | Not reported | Open |
| 28 | Grok 4.3xAI | 43.45% | Vals AIOriginal result | Not reported | Closed |
| 29 | Claude Sonnet 4.5 ThinkingAnthropic | 41.57% | Vals AIOriginal result | Not reported | Closed |
| 30 | MiniMax M2.5MiniMax | 41.57% | Vals AIOriginal result | Not reported | Closed |
| 31 | Qwen3.5 Plus ThinkingAlibaba | 41.57% | Vals AIOriginal result | Not reported | Unknown |
| 32 | Grok 4.20 0309 ReasoningxAI | 40.45% | Vals AIOriginal result | Not reported | Unknown |
| 33 | Kimi K2.5 ThinkingMoonshot AI | 40.45% | Vals AIOriginal result | Not reported | Unknown |
| 34 | GPT-5.4 nanoOpenAI | 39.89% | high reasoningOriginal result | Not reported | Closed |
| 35 | Gemma 4 31b ItGoogle | 39.33% | high reasoningOriginal result | Not reported | Unknown |
| 36 | Claude Haiku 4.5 ThinkingAnthropic | 38.20% | Vals AIOriginal result | Not reported | Closed |
| 37 | GLM-4.7Z.AI | 38.20% | Vals AIOriginal result | Not reported | Open |
| 38 | MiniMax M2.1MiniMax | 37.08% | Vals AIOriginal result | Not reported | Unknown |
| 39 | GPT-5OpenAI | 37.08% | high reasoningOriginal result | Not reported | Unknown |
| 40 | Kimi K2 ThinkingMoonshot AI | 37.08% | Vals AIOriginal result | Not reported | Unknown |
| 41 | DeepSeek V3p2 ThinkingFireworks AI | 35.95% | high reasoningOriginal result | Not reported | Unknown |
| 42 | DeepSeek V3p2Fireworks AI | 34.83% | none reasoningOriginal result | Not reported | Unknown |
| 43 | Laguna M.1Poolside | 31.46% | Vals AIOriginal result | Not reported | Closed |
| 44 | Gemini 2.5 ProGoogle | 30.34% | Vals AIOriginal result | Not reported | Closed |
| 45 | Mistral Medium 3.5Mistral AI | 30.34% | high reasoningOriginal result | Not reported | Unknown |
| 46 | Grok 4 Fast (Reasoning)xAI | 29.21% | Vals AIOriginal result | Not reported | Closed |
| 47 | Laguna XS.2Poolside | 28.09% | Vals AIOriginal result | Not reported | Open |
| 48 | GLM-4.6Z.AI | 28.09% | Vals AIOriginal result | Not reported | Open |
| 49 | Grok 4 0709xAI | 28.09% | Vals AIOriginal result | Not reported | Unknown |
| 50 | GPT-5 miniOpenAI | 26.97% | high reasoningOriginal result | Not reported | Closed |
| 51 | Moonshotai Kimi K2 InstructTogether AI | 25.84% | Vals AIOriginal result | Not reported | Unknown |
| 52 | Grok 4.1 Fast (Reasoning)xAI | 24.72% | Vals AIOriginal result | Not reported | Closed |
| 53 | Gemini 3.1 Flash Lite PreviewGoogle | 24.72% | high reasoningOriginal result | Not reported | Unknown |
| 54 | Qwen3.5 FlashAlibaba | 24.72% | Vals AIOriginal result | Not reported | Closed |
| 55 | Qwen3 MaxAlibaba | 24.72% | Vals AIOriginal result | Not reported | Closed |
| 56 | DeepSeek V3p1Fireworks AI | 22.47% | Vals AIOriginal result | Not reported | Unknown |
| 57 | Gemini 2.5 Flash Preview 09 2025 ThinkingGoogle | 21.35% | Vals AIOriginal result | Not reported | Unknown |
| 58 | Qwen3 MaxAlibaba | 20.23% | Vals AIOriginal result | Not reported | Closed |
| 59 | GPT-OSS 120BOpenAI | 19.10% | Vals AIOriginal result | Not reported | Open |
| 60 | Grok 4.1 Fast Non ReasoningxAI | 17.98% | Vals AIOriginal result | Not reported | Unknown |
| 61 | Trinity-Large-ThinkingArcee AI | 17.98% | Vals AIOriginal result | Not reported | Open |
| 62 | Mistral Small 2603Mistral AI | 16.85% | high reasoningOriginal result | Not reported | Unknown |
| 63 | GPT-4.1OpenAI | 14.61% | high reasoningOriginal result | Not reported | Closed |
| 64 | Magistral Medium 2509Mistral AI | 13.48% | Vals AIOriginal result | Not reported | Unknown |
| 65 | Mistral Large 2512Mistral AI | 8.99% | Vals AIOriginal result | Not reported | Unknown |
| 66 | Command A 03 2025Cohere | 2.25% | Vals AIOriginal result | Not reported | Unknown |
| 67 | Llama4 Maverick Instruct BasicFireworks AI | 2.25% | Vals AIOriginal result | Not reported | Unknown |
According to BenchLM.ai, GPT-5.5 leads the Terminal-Bench 2.0 benchmark with a score of 82%, followed by GPT-5.3 Codex (77.3%) and GPT-5.4 (75.1%). The scores show moderate spread, with meaningful differences between the top tier and mid-tier models.
45 models have been evaluated on Terminal-Bench 2.0. The benchmark falls in the Agentic category. BenchAlign v5.8 gives Terminal-Bench 2.0 7% 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 2.0
Year
2026
Tasks
Terminal-based software tasks
Format
Interactive CLI agent evaluation
Difficulty
Professional software engineering
Terminal-Bench 2.0 focuses on realistic CLI and repository workflows rather than toy code generation. It is a strong proxy for how useful a model is inside coding agents and autonomous developer tools.
Freshness and provenance
Version
Terminal-Bench 2
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
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 2.0 measure?
A benchmark for agentic software engineering tasks executed in real terminal environments. Models must inspect files, run commands, edit code, and recover from errors over multi-step workflows.
Which model scores highest on Terminal-Bench 2.0?
GPT-5.5 by OpenAI currently leads with a score of 82% on Terminal-Bench 2.0.
How many models are evaluated on Terminal-Bench 2.0?
45 AI models have published results on Terminal-Bench 2.0 in the BenchLM catalog.
Compare top models on Terminal-Bench 2.0
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.