Terminal-Bench-Science 0.1
We show this table for reference; we do not rank on it.
A benchmark of AI agents completing expert-curated research workflows across the life, physical, Earth, mathematical, and engineering sciences.
Resolution rate on Terminal-Bench-Science 0.1 — October 15, 2026 snapshot
We mirror the published resolution rate view for Terminal-Bench-Science 0.1. GPT-6 Astra leads the public snapshot at 68.1%, followed by Opus 5.5 (63.3%) and Claude Fable 5.1 (40.0%). We do not use these results to rank models overall.
GPT-6 Astra
OpenAI
Codex · max reasoning
Opus 5.5
Anthropic
Claude Code · max reasoning
Claude Fable 5.1
Anthropic
Claude Code · max reasoning
17 modelsAgenticCurrentDisplay onlyUpdated October 15, 2026 snapshot
Resolution rate table (17 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | GPT-6 AstraOpenAICodex · max reasoning | 68.1% | Not reported | Closed |
| 2 | Opus 5.5AnthropicClaude Code · max reasoning | 63.3% | Not reported | Unknown |
| 3 | Claude Fable 5.1AnthropicClaude Code · max reasoning | 40.0% | Not reported | Closed |
| 4 | Claude Opus 5AnthropicClaude Code · max reasoning | 30.0% | Not reported | Closed |
| 5 | GPT-5.6 SolOpenAICodex · max reasoning | 22.4% | Not reported | Closed |
| 6 | Claude Fable 5AnthropicClaude Code · max reasoning | 21.4% | Not reported | Closed |
| 7 | DeepSeek V4.1 Flash 1017DeepSeekCodex · max reasoning | 15.7% | Not reported | Open |
| 8 | Grok 4.7xAIGrok Build · xhigh reasoning | 14.3% | Not reported | Closed |
| 9 | Gemini 3.8 FlashGooglemini-SWE-agent · high reasoning | 12.4% | Not reported | Closed |
| 10 | Claude Opus 4.8AnthropicClaude Code · max reasoning | 10.5% | Not reported | Closed |
| 11 | GPT-5.6 TerraOpenAICodex · max reasoning | 8.6% | Not reported | Closed |
| 12 | GLM-5.3Z.AIClaude Code · max reasoning | 8.1% | Not reported | Open |
| 13 | Kimi K3Moonshot AIClaude Code · max reasoning | 7.1% | Not reported | Pending |
| 14 | Grok 4.6xAIGrok Build · xhigh reasoning | 7.1% | Not reported | Closed |
| 15 | Gemini 3.7 FlashGooglemini-SWE-agent · high reasoning | 5.7% | Not reported | Closed |
| 16 | DeepSeek V4 Pro 0813DeepSeekCodex · max reasoning | 3.8% | Not reported | Open |
| 17 | GPT-5.6 LunaOpenAICodex · max reasoning | 3.3% | Not reported | Closed |
Terminal-Bench Science (Vals)
We keep the Vals fixed-Terminus-2, single-trial 70-task run separate from the official Terminal-Bench Science 0.1 leaderboard, which uses native agents and three trials per task. This mirror is display only.
| Source rank | Source model or configuration | 70-task pass@1 | Source setup | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | GPT-6 AstraOpenAI | 62.86% | Mini-SWE-agent · max reasoningOriginal result | Not reported | Closed |
| 2 | GPT-6.1 SolOpenAI | 52.86% | Mini-SWE-agent · max reasoningOriginal result | Not reported | Closed |
| 3 | Claude Opus 5.5Anthropic | 47.14% | Mini-SWE-agentOriginal result | Not reported | Closed |
| 4 | Claude Sonnet 5.5Anthropic | 47.14% | Mini-SWE-agentOriginal result | Not reported | Closed |
| 5 | Gemini 4 ArgonGoogle | 44.29% | Mini-SWE-agent · high reasoningOriginal result | Not reported | Closed |
| 6 | Claude Fable 5.1Anthropic | 40.00% | Mini-SWE-agentOriginal result | Not reported | Closed |
| 7 | Muse Wonder FullMeta | 37.14% | Mini-SWE-agent · max reasoningOriginal result | Not reported | Closed |
| 8 | GPT-6 SolOpenAI | 30.00% | Mini-SWE-agent · max reasoningOriginal result | Not reported | Closed |
| 9 | Claude Opus 5Anthropic | 27.14% | Mini-SWE-agentOriginal result | Not reported | Closed |
| 10 | Muse WonderMeta | 22.86% | Mini-SWE-agent · max reasoningOriginal result | Not reported | Closed |
| 11 | GPT-5.6 SolOpenAI | 20.00% | Mini-SWE-agent · max reasoningOriginal result | Not reported | Closed |
| 12 | Claude Fable 5Anthropic | 15.71% | Mini-SWE-agentOriginal result | Not reported | Closed |
| 13 | Qwen3.8 MaxAlibaba | 12.86% | Mini-SWE-agentOriginal result | Not reported | Open |
| 14 | Grok 4.6xAI | 11.43% | Mini-SWE-agent · high reasoningOriginal result | Not reported | Closed |
| 15 | Muse Spark 1.3 MaxMeta | 10.00% | Mini-SWE-agent · max reasoningOriginal result | Not reported | Unknown |
| 16 | Grok 4.7xAI | 10.00% | Mini-SWE-agent · xhigh reasoningOriginal result | Not reported | Closed |
| 17 | Claude Haiku 5.5Anthropic | 10.00% | Mini-SWE-agentOriginal result | Not reported | Closed |
| 18 | GPT-5.6 TerraOpenAI | 10.00% | Mini-SWE-agent · max reasoningOriginal result | Not reported | Closed |
| 19 | Gemini 3.8 FlashGoogle | 8.57% | Mini-SWE-agent · high reasoningOriginal result | Not reported | Closed |
| 20 | Gemini 3.7 FlashGoogle | 7.14% | Mini-SWE-agent · high reasoningOriginal result | Not reported | Closed |
| 21 | Gemini 3.5 FlashGoogle | 5.71% | Mini-SWE-agent · high reasoningOriginal result | Not reported | Closed |
| 22 | GLM-5.3Z.AI | 5.71% | Mini-SWE-agent · max reasoningOriginal result | Not reported | Open |
| 23 | Claude Sonnet 5Anthropic | 5.71% | Mini-SWE-agentOriginal result | Not reported | Closed |
| 24 | Ember-1Fireworks | 5.71% | Mini-SWE-agentOriginal result | Not reported | Closed |
| 25 | Muse Spark 1.3Meta | 4.29% | Mini-SWE-agent · xhigh reasoningOriginal result | Not reported | Closed |
| 26 | GPT-6 LunaOpenAI | 4.29% | Mini-SWE-agent · max reasoningOriginal result | Not reported | Closed |
| 27 | DeepSeek V4 Flash 0731DeepSeek | 4.29% | Mini-SWE-agent · high reasoningOriginal result | Not reported | Open |
| 28 | Gemini 3.6 FlashGoogle | 4.29% | Mini-SWE-agent · high reasoningOriginal result | Not reported | Closed |
| 29 | Claude Opus 4.8Anthropic | 4.29% | Mini-SWE-agentOriginal result | Not reported | Closed |
| 30 | DeepSeek V4 Pro 0813DeepSeek | 4.29% | Mini-SWE-agent · max reasoningOriginal result | Not reported | Open |
| 31 | MiMo-V2.6-FlashXiaomi | 4.29% | Mini-SWE-agentOriginal result | Not reported | Open |
| 32 | Step 5 PreviewStepFun | 2.86% | Mini-SWE-agentOriginal result | Not reported | Open |
| 33 | MiMo-V2.6-ProXiaomi | 2.86% | Mini-SWE-agentOriginal result | Not reported | Open |
| 34 | GLM-5.3-FlashZ.AI | 2.86% | Mini-SWE-agent · max reasoningOriginal result | Not reported | Open |
| 35 | Gemini 3.1 Pro PreviewGoogle | 1.43% | Mini-SWE-agent · high reasoningOriginal result | Not reported | Unknown |
| 36 | Hy4 previewTencent | 1.43% | Terminus 2Original result | Not reported | Open |
| 37 | Kimi K3Moonshot AI | 1.43% | Mini-SWE-agent · max reasoningOriginal result | Not reported | Pending |
| 38 | Mercury 2.5Inception | 0.00% | Mini-SWE-agent · high reasoningOriginal result | Not reported | Closed |
| 39 | GPT-5.6 LunaOpenAI | 0.00% | Mini-SWE-agent · max reasoningOriginal result | Not reported | Closed |
| 40 | DeepSeek V4.1 Flash 1017DeepSeek | 0.00% | Mini-SWE-agent · high reasoningOriginal result | Not reported | Open |
| 41 | Qwen3.8-27BAlibaba | 0.00% | Mini-SWE-agent · xhigh reasoningOriginal result | Not reported | Open |
How to read this leaderboard
Operator receipt: 17 sourced rows are currently displayable on this page; the leading published row is GPT-6 Astra at 68.1%.
Honest limit: Each result combines a model, agent harness, and science-specific environment. The 70-task release is deliberately difficult and continuously maintained, so read the table as a versioned system evaluation rather than a base-model capability score.
How we show Terminal-Bench-Science 0.1
We mirror the official Terminal-Bench-Science 0.1 leaderboard across 70 expert-curated research workflows in the life, physical, Earth, mathematical, and engineering sciences. Each system runs 3 independent trials per task and is graded on concrete artifacts such as analyses, simulations, proofs, code, and data products.
Each row combines a model with an agent harness. The benchmark has its own task set and protocol, so we keep it as a standalone display-only surface outside the external-consensus scoring pipeline.
Snapshot
The published Terminal-Bench-Science 0.1 snapshot places GPT-6 Astra first at 68.1%. The third row is 28.1 points behind. The broader top-10 range is 57.6 points, so the table still separates the published systems.
17 models have been evaluated on Terminal-Bench-Science 0.1. The benchmark falls in the Agentic category. Terminal-Bench-Science 0.1 is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Terminal-Bench-Science 0.1
Year
2026
Tasks
70 expert-curated scientific research workflows
Format
Resolution rate across 3 trials per task
Difficulty
Frontier scientific research workflows
The first release contains 70 workflows contributed and reviewed by researchers across five scientific domains. Each model-and-agent system runs three independent trials per task and produces concrete artifacts graded by reproducible, task-specific tests. We mirror the official system leaderboard as a standalone display-only benchmark.
Freshness and provenance
Version
Terminal-Bench-Science 0.1.0
Refresh cadence
Continuous
Staleness state
Current
Question availability
70 public task definitions
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-Science 0.1 measure?
A benchmark of AI agents completing expert-curated research workflows across the life, physical, Earth, mathematical, and engineering sciences.
Which model leads the published Terminal-Bench-Science 0.1 snapshot?
GPT-6 Astra currently leads the published Terminal-Bench-Science 0.1 snapshot with 68.1% resolution rate. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on Terminal-Bench-Science 0.1?
The October 15, 2026 snapshot snapshot contains 17 AI models.
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 5,600+ readers.
One email each week. Unsubscribe anytime.