BlogLLM Benchmark Scores as Published, Side by Side
Terminal-Bench Science: Results as Published
Every result the catalog holds for Terminal-Bench Science: 15 models from 6 developers, read from 11 pages. Each number is as printed by the company named beside it; nothing is ranked or rescored. Models are listed by developer, newest first.
Area Science and MathMaintainer Terminal-Bench-Science team, led by researchers at Stanford University, built with the Terminal-Bench and Harbor teamIntroduced 2026Data version 2026103001, as of 2026-10-30
Terminal-Bench-Science tests AI agents on the computational workflows working scientists actually run, not textbook questions. Each task is a containerised terminal job, such as statistical analysis, simulation, signal processing or proofs, and the output is checked by task-specific automated tests. Version 0.1 has 70 tasks written by practising researchers across life, physical, Earth, mathematical and engineering sciences.
Terminal-Bench Science 0.1
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-6 AstraOpenAI, released 2026-09-03 | 64.6% | OpenAI, GPT-6 Astrachecked 2026-10-14; printed on 7 more pages | - |
| GPT-5.6 SolOpenAI, released 2026-07-09 | 22.4% | OpenAI, GPT-6 Astrachecked 2026-10-14; printed on 3 more pages | - |
| Claude Haiku 5.5Anthropic, released 2026-10-07 | 20.6% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
| Claude Sonnet 5.5Anthropic, released 2026-09-28 | 59.9% | Anthropic, Claude Sonnet 5.5 System Cardchecked 2026-10-19; printed on 1 more page | Effort: max; Harness: Claude Code --bare; Fallback: No fallback fired on any request or trial |
| Claude Opus 5.5Anthropic, released 2026-09-22 | 58.7% | Anthropic, Claude Opus 5.5checked 2026-10-14; printed on 6 more pages | - |
| Claude Fable 5.1Anthropic, released 2026-09-01 | 52.6% | Anthropic, Claude Fable 5.1 and Mythos 5.1checked 2026-10-15; printed on 8 more pages | - |
| Claude Opus 5Anthropic, released 2026-07-24 | 29% | Anthropic, Claude Opus 5.5checked 2026-10-14; printed on 6 more pages | - |
| Claude Fable 5Anthropic, released 2026-06-09 | 24.7% | Anthropic, Claude Fable 5.1 and Mythos 5.1checked 2026-10-15; printed on 5 more pages | - |
| Claude Haiku 4.5Anthropic, released 2025-10-15 | 0.1% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
| Gemini 4 CarbonGoogle, released 2026-10-28 | 64.8% | Google, Gemini 4 Carbonchecked 2026-10-29 | - |
| Gemini 4 ArgonGoogle, released 2026-09-30 | 57.6% | Google, Gemini 4 Argonchecked 2026-10-29; printed on 1 more page | - |
| DeepSeek-V4.1-Flash-1017DeepSeek, released 2026-10-17 | 9.3% | Moonshot AI, Introducing Kimi K3.1 Prochecked 2026-10-24 | Effort: max; Model as printed: DeepSeek-V4.1-Flash-1017 |
| 15.7% | Moonshot AI, Introducing Kimi K3.1 Prochecked 2026-10-24 | Effort: max; Model as printed: DeepSeek-V4.1-Flash-1017 (max) | |
| Kimi K3.1 ProMoonshot AI, released 2026-10-23 | 38.1% | Moonshot AI, Introducing Kimi K3.1 Prochecked 2026-10-24 | Effort: max; Harness: Claude Code |
| GLM-5.3Z.ai, released 2026-08-18 | 8.1% | Moonshot AI, Introducing Kimi K3.1 Prochecked 2026-10-24 | Effort: max |
Terminal-Bench Science 0.1 (average cost per task)
lower is better
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-6.1 SolOpenAI, released 2026-09-29 | $5.47 | OpenAI, GPT-6.1 Solchecked 2026-10-14 | - |
| GPT-6 AstraOpenAI, released 2026-09-03 | $23.8 | OpenAI, GPT-6.1 Solchecked 2026-10-14 | - |
| Claude Opus 5.5Anthropic, released 2026-09-22 | $23.21 | OpenAI, GPT-6.1 Solchecked 2026-10-14 | - |
Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.