Terminal-Bench
Agentic terminal tasks. Epoch publishes one row per agent harness; the best score per model is kept, because the harness is not a property of the model. Aggregated from Epoch AI's AI Benchmarking Hub (terminalbench_external.csv, column 'Accuracy mean'), used under CC-BY 4.0. The Known Good aggregates published results; it runs no evaluations.
52 models scored
Higher is better
Published by Epoch AI
CC-BY 4.0
Terminal-Bench
Terminal-Bench as measured and published by Epoch AI (CC-BY 4.0). Ingested, normalised only where it feeds an index, and never re-run by us. Higher is better.
💡 A lightbulb marks a model its provider publishes as a reasoning model
The Known Good
Every model we hold a Terminal-Bench score for
All 52 of 52 models
| # | Model | Creator | Terminal-Bench | Known Good Index | Measured |
|---|---|---|---|---|---|
| 1 |
OpenAI: GPT-5.5
💡 |
OpenAI | 84.7 | 93 | 4 Sep 2026 |
| 2 |
OpenAI: GPT-5.4
💡 |
OpenAI | 81.8 | 91 | 4 Sep 2026 |
| 3 |
Anthropic: Claude Opus 4.7
💡 |
Anthropic | 80.2 | 92 | 4 Sep 2026 |
| 4 |
Google: Gemini 3.1 Pro Preview
💡 |
80.2 | 97 | 4 Sep 2026 | |
| 5 |
Anthropic: Claude Opus 4.6
💡 |
Anthropic | 79.8 | 89 | 4 Sep 2026 |
| 6 |
OpenAI: GPT-5.3-Codex
💡 |
OpenAI | 78.4 | — | 4 Sep 2026 |
| 7 |
Gemini 3 Pro
|
69.4 | — | 4 Sep 2026 | |
| 8 |
OpenAI: GPT-5.2-Codex
💡 |
OpenAI | 66.5 | — | 4 Sep 2026 |
| 9 |
OpenAI: GPT-5.2
💡 |
OpenAI | 64.9 | 81 | 4 Sep 2026 |
| 10 |
gemini-3-flash
|
64.3 | — | 4 Sep 2026 | |
| 11 |
Anthropic: Claude Opus 4.5
💡 |
Anthropic | 63.1 | 72 | 4 Sep 2026 |
| 12 |
gemini-3-pro-preview
|
61.8 | 84 | 25 Aug 2026 | |
| 13 |
OpenAI: GPT-5.1-Codex-Mini
💡 |
OpenAI | 61.6 | — | 4 Sep 2026 |
| 14 |
OpenAI: GPT-5.1-Codex-Max
💡 |
OpenAI | 60.4 | — | 4 Sep 2026 |
| 15 |
claude-opus-4-5-20251101_128K
|
Anthropic | 59.1 | — | 4 Sep 2026 |
| 16 |
OpenAI: GPT-5.1-Codex
💡 |
OpenAI | 57.8 | — | 4 Sep 2026 |
| 17 |
SpaceXAI: Grok 4.20
💡 |
xAI | 57.3 | — | 4 Sep 2026 |
| 18 |
Anthropic: Claude Sonnet 4.6
💡 |
Anthropic | 53.4 | 80 | 4 Sep 2026 |
| 19 |
Z.ai: GLM 5
💡 |
Z.ai | 52.4 | 77 | 4 Sep 2026 |
| 20 |
Google: Gemini 3 Flash Preview
💡 |
51.0 | 83 | 4 Sep 2026 | |
| 21 |
OpenAI: GPT-5
💡 |
OpenAI | 49.6 | 78 | 4 Sep 2026 |
| 22 |
OpenAI: GPT-5.1
💡 |
OpenAI | 47.6 | 74 | 4 Sep 2026 |
| 23 |
Anthropic: Claude Sonnet 4.5
💡 |
Anthropic | 46.5 | — | 4 Sep 2026 |
| 24 |
MiniMax: MiniMax M2.7
💡 |
MiniMax | 45.1 | — | 4 Sep 2026 |
| 25 |
OpenAI: GPT-5 Codex
💡 |
OpenAI | 44.3 | — | 4 Sep 2026 |
| 26 |
MoonshotAI: Kimi K2.5
💡 |
Moonshot AI | 43.2 | — | 4 Sep 2026 |
| 27 |
claude-sonnet-4-5-20250929
|
Anthropic | 42.8 | 66 | 4 Sep 2026 |
| 28 |
MiniMax: MiniMax M2.5
💡 |
MiniMax | 42.7 | — | 4 Sep 2026 |
| 29 |
Claude 4.5 Sonnet
|
Anthropic | 42.7 | — | 4 Sep 2026 |
| 30 |
DeepSeek: DeepSeek V3.2
💡 |
DeepSeek | 39.6 | — | 4 Sep 2026 |
| 31 |
Anthropic: Claude Opus 4.1
💡 |
Anthropic | 38.0 | 50 | 4 Sep 2026 |
| 32 |
MiniMax: MiniMax M2.1
💡 |
MiniMax | 36.6 | — | 4 Sep 2026 |
| 33 |
MoonshotAI: Kimi K2 Thinking
💡 |
Moonshot AI | 35.7 | — | 4 Sep 2026 |
| 34 |
Anthropic: Claude Haiku 4.5
💡 |
Anthropic | 35.5 | — | 4 Sep 2026 |
| 35 |
OpenAI: GPT-5 Mini
💡 |
OpenAI | 34.8 | 67 | 4 Sep 2026 |
| 36 |
Z.ai: GLM 4.7
💡 |
Z.ai | 33.4 | 69 | 4 Sep 2026 |
| 37 |
Google: Gemini 2.5 Pro
💡 |
32.6 | 65 | 4 Sep 2026 | |
| 38 |
MiniMax: MiniMax M2
💡 |
MiniMax | 30.0 | — | 4 Sep 2026 |
| 39 |
claude-haiku-4-5-20251001
|
Anthropic | 29.8 | 67 | 4 Sep 2026 |
| 40 |
Kimi-K2-Instruct
|
Moonshot AI | 27.8 | — | 4 Sep 2026 |
| 41 |
grok-4-0709
|
xAI | 27.2 | 68 | 4 Sep 2026 |
| 42 |
Qwen 3 Coder 480B
|
Alibaba | 27.2 | — | 4 Sep 2026 |
| 43 |
grok-code-fast-1
|
xAI | 25.8 | — | 4 Sep 2026 |
| 44 |
Qwen3-Coder-480B-A35B-Instruct
|
Alibaba | 25.4 | — | 4 Sep 2026 |
| 45 |
Z.ai: GLM 4.6
💡 |
Z.ai | 24.5 | — | 4 Sep 2026 |
| 46 |
Qwen: Qwen3.6 35B A3B
💡 |
Alibaba | 23.0 | 66 | 4 Sep 2026 |
| 47 |
OpenAI: GPT-5 Nano
💡 |
OpenAI | 21.8 | 68 | 4 Sep 2026 |
| 48 |
OpenAI: gpt-oss-120b
💡 |
OpenAI | 18.7 | 62 | 4 Sep 2026 |
| 49 |
Google: Gemini 2.5 Flash
💡 |
17.1 | — | 4 Sep 2026 | |
| 50 |
gemini-2.5-flash-preview-09-2025
|
17.1 | — | 4 Sep 2026 | |
| 51 |
Qwen: Qwen3.5-9B
💡 |
Alibaba | 9.2 | 44 | 4 Sep 2026 |
| 52 |
OpenAI: gpt-oss-20b
💡 |
OpenAI | 3.4 | 42 | 4 Sep 2026 |
The Known Good
Provenance. These figures are published by Epoch AI and redistributed here under CC-BY 4.0.
The publisher's own page is here. We did not run this evaluation and we do not adjust the published numbers — where a score feeds an index it is
min-max normalised against every other model holding the same evaluation, and nothing else is done to it.
This evaluation is one of the seven components of the
Known Good Index. Full licence detail is on attribution, and every score here is
in the CSV download.