Why this benchmark is useful
Editorial brief pendingWe publish the methodology and ledger first; benchmark-specific analysis ships after desk review.
Benchmarks / Agentic
Terminal-Bench
Terminal-Bench 2.1
Whether an agent can complete realistic, containerized terminal tasks (Terminal-Bench 2.1): debugging code, handling files, running commands, and producing the artifact the hidden checker expects. 16% weight in AA Intelligence Index v4.1.
AgenticSolidActiveLow contamination riskSince 2025
What this does not measure
- The model alone — harness, tools, retries, time limits, and container setup can move scores dramatically.
- Code maintainability — a passed task can still produce brittle scripts or unreviewable patches.
- Every developer workflow — terminal tasks are closer to agentic operations than product design, frontend polish, or long-running repository stewardship.
Coverage map
- Task family
- Agentic
- Format
- Containerized tasks with a terminal environment; agents operate through commands and are graded by task-specific checks.
- Scoring
- Resolution/pass rate across audited tasks. Always read the agent harness and model pairing.
- Maintainer
- Laude Institute / Terminal-Bench
How to read the scores
Reading guide pending. Use task format, scoring method, and source dates in the ledger until the desk brief ships.
What it does not cover
- The model alone — harness, tools, retries, time limits, and container setup can move scores dramatically.
- Code maintainability — a passed task can still produce brittle scripts or unreviewable patches.
- Every developer workflow — terminal tasks are closer to agentic operations than product design, frontend polish, or long-running repository stewardship.
Scores
69 rowsEvidence ledger
6969 rows
88.01%best score
68source-checked
3sources
2026-07-21to 2026-05-20
Where the rows land
0255075100
Normalized to this benchmark's axis (0–100). Open the table for raw units.
Newest receipts
- GPT-5.6 Sol88.01% ·
- GPT-5.6 Terra88.01% ·
- Kimi K385.02% ·
- Claude Fable 584.64% ·
- Grok 4.581.65% ·
Scores use this benchmark's own unit and axis, not a universal quality score.
| # | Model | Release date | Score | Provenance | Trust |
|---|---|---|---|---|---|
| 1 | GPT-5.6 SolOpenAI | 2026-07-09 | 88.01% | 2026-07-21 | Source-checked |
| 2 | GPT-5.6 TerraOpenAI | 2026-07-09 | 88.01% | 2026-07-21 | Source-checked |
| 3 | Kimi K3Moonshot | 2026-07-16 | 85.02% | 2026-07-21 | Source-checked |
| 4 | Claude Fable 5Anthropic | 2026-06-09 | 84.64% | 2026-07-21 | Source-checked |
| 1 | Claude Mythos PreviewAnthropic | 2026-06-01 | 84.1% | Terminal-Bench 2.0 leaderboard2026-06-02 | Source-checked |
| 2 | GPT 5.5 (Codex CLI)OpenAI | 2026-04-23 | 82% | Terminal-Bench 2.0 leaderboard2026-05-20 | Source-checked |
| 3 | GPT 5.4 (ForgeCode)OpenAI | 2026-03-05 | 81.8% | Terminal-Bench 2.0 leaderboard2026-05-20 | Source-checked |
| 8 | Grok 4.5xAI | 2026-07-08 | 81.65% | 2026-07-21 | Source-checked |
| 9 | GPT-5.6 LunaOpenAI | 2026-07-09 | 80.9% | 2026-07-21 | Source-checked |
| 10 | Claude Sonnet 5Anthropic | 2026-06-30 | 80.52% | 2026-07-21 | Source-checked |
| 11 | GPT-5.5 (medium)OpenAI | 2026-04-23 | 80.52% | 2026-07-21 | Source-checked |
| 4 | Claude Opus 4.6 (ForgeCode)Anthropic | 2026-02-01 | 79.8% | Terminal-Bench 2.0 leaderboard2026-05-20 | Source-checked |
| 13 | GPT-5.5 (high)OpenAI | 2026-04-23 | 79.4% | 2026-07-21 | Source-checked |
| 14 | GLM-5.2Zhipu | 2026-06-13 | 77.9% | 2026-07-21 | Source-checked |
| 15 | Muse Spark 1.1Meta | 2026-07-09 | 77.9% | 2026-07-21 | Source-checked |
| 4 | Gemini 3.5 FlashGoogle | 2026-05-19 | 76.2% | Terminal-Bench 2.0 leaderboard2026-05-20 | Source-checked |
| 5 | Claude Opus 4.8Anthropic | 2026-05-28 | 74.6% | Terminal-Bench 2.0 leaderboard2026-05-20 | Source-checked |
| 18 | Qwen3.7 MaxAlibaba | 2026-05-19 | 74.53% | 2026-07-21 | Source-checked |
| 19 | Gemini 3.1 Pro PreviewGoogle | 2026-02-19 | 73.78% | 2026-07-21 | Source-checked |
| 20 | Claude Sonnet 4.6 (Adaptive Reasoning, Max Effort)Anthropic | 2026-02-17 | 71.16% | 2026-07-21 | Source-checked |
| 21 | LongCat-2.0MeituanVendor-reported LongCat-2.0 score; keep separate from audited Terminal-Bench leaderboard rows until an independent run lands. | 2026-06-29 | 70.8% | Meituan LongCat-2.0 release materials2026-06-29 | Needs audit |
| 6 | Qwen 3.7 MaxAlibaba | 2026-05-20 | 69.7% | Terminal-Bench 2.0 leaderboard2026-05-20 | Source-checked |
| 7 | Claude Opus 4.7 (Adaptive)Anthropic | 2026-04-16 | 69.4% | Terminal-Bench 2.0 leaderboard2026-05-20 | Source-checked |
| 24 | Claude Sonnet 4.6Anthropic | 2026-02-17 | 68.4% | Terminal-Bench 2.0 leaderboard2026-06-09 | Source-checked |
| 8 | DeepSeek V4 Pro MaxDeepSeek | 2026-04-24 | 67.9% | Terminal-Bench 2.0 leaderboard2026-05-20 | Source-checked |
| 26 | Kimi K2.7 CodeMoonshot | 2026-06-12 | 67.42% | 2026-07-21 | Source-checked |
| 9 | Kimi K2.6Moonshot | 2026-04-20 | 66.7% | Terminal-Bench 2.0 leaderboard2026-05-20 | Source-checked |
| 10 | MiniMax M3MiniMax | 2026-05-31 | 66% | Terminal-Bench 2.0 leaderboard2026-05-20 | Source-checked |
| 29 | GPT-5.5 (low)OpenAI | 2026-04-23 | 65.54% | 2026-07-21 | Source-checked |
| 11 | Gemini 3.1 ProGoogle | 2026-02-19 | 65.4% | Terminal-Bench 2.0 leaderboard2026-05-20 | Source-checked |
| 31 | MiMo-V2.5-ProXiaomi | 2026-04-22 | 65.17% | 2026-07-21 | Source-checked |
| 32 | DeepSeek V4 Pro (Reasoning, Max Effort)DeepSeek | 2026-04-24 | 64.04% | 2026-07-21 | Source-checked |
| 33 | Grok 4.3xAI | 2026-04-30 | 63.8% | Terminal-Bench 2.0 leaderboard2026-06-09 | Source-checked |
| 34 | Muse SparkOther | 2026-01-01 | 62.17% | 2026-07-21 | Source-checked |
| 35 | GLM-5.1Zhipu | 2026-04-07 | 61.8% | 2026-07-21 | Source-checked |
| 36 | Claude Haiku 4.5Anthropic | 2025-10-01 | 61.4% | Terminal-Bench 2.0 leaderboard2026-06-09 | Source-checked |
| 37 | Qwen 3.7 PlusAlibaba | 2026-06-02 | 61.05% | 2026-07-21 | Source-checked |
| 38 | GPT-5.4 NanoOpenAI | 2026-03-17 | 60.67% | 2026-07-21 | Source-checked |
| 39 | GPT-5.4 MiniOpenAI | 2026-03-17 | 59.18% | 2026-07-21 | Source-checked |
| 40 | Llama 4 ScoutMeta | 2026-04-05 | 58.6% | Terminal-Bench 2.0 leaderboard2026-06-09 | Source-checked |
| 41 | MiniMax M2.7MiniMax | 2026-04-15 | 55.43% | 2026-07-21 | Source-checked |
| 42 | Claude Opus 4.7 (Non-reasoning, High Effort)Anthropic | 2026-04-16 | 54.55% | 2026-07-21 | Source-checked |
| 43 | Nemotron 3 Ultra 550B A55B (Reasoning)NVIDIA | 2026-06-04 | 53.93% | 2026-07-21 | Source-checked |
| 44 | GPT-5.3 Codex (xhigh)OpenAI | 2026-02-05 | 53.03% | 2026-07-21 | Source-checked |
| 45 | GPT-5.1OpenAI | 2025-11-13 | 52.43% | 2026-07-21 | Source-checked |
| 46 | GPT-5.2OpenAI | 2025-12-11 | 46.97% | 2026-07-21 | Source-checked |
| 47 | Claude Opus 4.6 (Adaptive Reasoning, Max Effort)Anthropic | 2026-02-05 | 46.21% | 2026-07-21 | Source-checked |
| 48 | Kimi K2.5Moonshot | 2026-01-27 | 45.69% | 2026-07-21 | Source-checked |
| 49 | Gemma 4 31B ITGoogle | 2026-03-01 | 43.45% | 2026-07-21 | Source-checked |
| 50 | Gemini 3 ProGoogle | 2025-11-18 | 41.67% | 2026-07-21 | Source-checked |
| 51 | Claude Opus 4.5Anthropic | 2025-11-24 | 40.91% | 2026-07-21 | Source-checked |
| 52 | Gemini 3.5 Flash (medium)Google | 2026-05-19 | 39.39% | 2026-07-21 | Source-checked |
| 53 | Step 3.7 FlashStepFun | 2026-06-01 | 39.33% | 2026-07-21 | Source-checked |
| 54 | Grok 4.20 ReasoningxAI | 2026-03-05 | 37.88% | 2026-07-21 | Source-checked |
| 55 | OpenAI o3OpenAI | 2025-04-16 | 37.12% | 2026-07-21 | Source-checked |
| 56 | North Mini CodeCohere | 2026-06-09 | 35.58% | 2026-07-21 | Source-checked |
| 57 | MiniMax M2.5MiniMax | 2026-04-01 | 34.85% | 2026-07-21 | Source-checked |
| 58 | Gemini 3 Flash PreviewGoogle | 2025-12-17 | 31.82% | 2026-07-21 | Source-checked |
| 59 | Gemini 3.1 Flash LiteGoogle | 2026-05-07 | 31.09% | 2026-07-21 | Source-checked |
| 60 | HyperNova 60B 2605Multiverse | 2026-05-06 | 18.35% | 2026-07-21 | Source-checked |
| 61 | Gemma 4 12B (Reasoning)Google | 2026-06-07 | 18.18% | 2026-07-21 | Source-checked |
| 62 | DeepSeek R1DeepSeek | 2025-01-20 | 15.91% | 2026-07-21 | Source-checked |
| 63 | OpenAI o1OpenAI | 2024-09-12 | 12.88% | 2026-07-21 | Source-checked |
| 64 | Mistral Large 3Mistral | 2025-12-02 | 11.99% | 2026-07-21 | Source-checked |
| 65 | Gemma 4 12B (Non-reasoning)Google | 2026-06-10 | 11.36% | 2026-07-21 | Source-checked |
| 66 | Llama 4 MaverickMeta | 2026-04-05 | 7.87% | 2026-07-21 | Source-checked |
| 67 | LFM2.5-8B-A1BLiquid | 2026-06-08 | 4.55% | 2026-07-21 | Source-checked |
| 68 | Command ACohere | 2026-03-15 | 0.76% | 2026-07-21 | Source-checked |
| 69 | MiniCPM5-1B (Reasoning)OpenBMB | 2026-06-04 | 0% | 2026-07-21 | Source-checked |
What it covers
Data-quality note pending. Every ledger row still carries source URL, source date, and ingest timestamp.
Where it breaks down
No ceiling note is recorded yet. Treat clustering near the top as a warning that the benchmark may no longer separate frontier models.
Receipts
Sources and further reading
- Official Terminal-Bench leaderboard· checked 2026-06-03
- Terminal-Bench 2 repository· checked 2026-06-03
- Also tracked on Vals AI (independent run)· checked 2026-06-09
- Leaderboard
The Pack · Editorial newsletter
New cards in your inbox. Free.
One short email when a card ships or changes status. No tracking, no third-party analytics. Unsubscribe in one click.