Terminal-Bench 4.0 leaderboard (Oct 2026): AI model scores
As of 19 Oct 2026, the highest reported Anthropic score on Terminal-Bench 4.0 is Claude Opus 5.5 (Anthropic) at 65.15%, #1 on that leaderboard, per Vals AI.
What it measures
Frontier-difficulty terminal tasks across software, science, ML, operations, hardware, security and media.
Hard, multi-step tasks an agent must complete on its own in a command-line environment.
What the score means
Score: pass@1, averaged over 3 runs
Share of terminal tasks completed and verified by tests.
Results from published AI leaderboards. Compare models within each benchmark; scores from different tests aren’t directly comparable. How we collect these results
Scores by model
Run by Vals AI (benchmark by Terminal-Bench team). Leaderboard dated 19 Oct 2026. Last checked 22 Oct 2026. Showing top 25 of 52 models listed.
Highlight a provider: Anthropic OpenAI Google xAI Meta DeepSeek Alibaba Z.ai MiniMax Mistral Xiaomi
Highlighting Anthropic models. Other providers are dimmed. Show all providers
- Anthropic
- OpenAI
- xAI
- Meta
- DeepSeek
- Alibaba
- Z.ai
- MiniMax
- Mistral
- Xiaomi
- Other
- Score
- Share of terminal tasks completed and verified by tests.
- Cost per test (USD)
- Average API cost per task, as published.
All scores
| Rank | Model | Provider | Score | Cost per test (USD) | Harness | Price in / out ($/1M tokens) |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | Anthropic | 65.15% | $13 | Mini-SWE-agent | $4 / $20 |
| 2 | Claude Sonnet 5.5 | Anthropic | 64.14% | $17 | Mini-SWE-agent | $2 / $10 |
| 3 | GPT-6 Astra | OpenAI | 59.60% | $9.58 | Mini-SWE-agent | $10 / $50 |
| 4 | Claude Fable 5.1 | Anthropic | 58.08% | $17 | Mini-SWE-agent | $10 / $50 |
| 5 | Wonder Full | Meta | 58.08% | $11 | Mini-SWE-agent | $4 / $20 |
| 6 | Gemini 4 Argon | 57.58% | $18 | Mini-SWE-agent | $4 / $20 | |
| 7 | GPT-6.1 Sol | OpenAI | 55.05% | $1.72 | Mini-SWE-agent | $2 / $10 |
| 8 | Claude Opus 5 | Anthropic | 53.53% | $19 | Mini-SWE-agent | $5 / $25 |
| 9 | Wonder | Meta | 49.49% | $15 | Mini-SWE-agent | $4 / $20 |
| 10 | GPT-6 Sol | OpenAI | 44.44% | $5.79 | Mini-SWE-agent | $2 / $10 |
Show 15 more rows
| Rank | Model | Provider | Score | Cost per test (USD) | Harness | Price in / out ($/1M tokens) |
|---|---|---|---|---|---|---|
| 11 | Claude Fable 5 | Anthropic | 41.41% | $30 | Mini-SWE-agent | $10 / $50 |
| 12 | GLM 5.3 | Z.ai | 38.89% | $9.37 | Mini-SWE-agent | $1.4 / $4.4 |
| 13 | GPT-5.6 Sol | OpenAI | 37.88% | $7.98 | Mini-SWE-agent | $4 / $20 |
| 14 | GLM 5.3 Flash | Z.ai | 37.37% | $1.22 | Mini-SWE-agent | $0.075 / $0.25 |
| 15 | Claude Haiku 5.5 | Anthropic | 36.36% | $3.12 | Mini-SWE-agent | $0.1 / $0.5 |
| 16 | Qwen 3.8 Max | Alibaba | 34.34% | $11 | Mini-SWE-agent | $2 / $6 |
| 17 | DeepSeek V4.1 Flash 1017 | DeepSeek | 31.82% | $0.54 | Mini-SWE-agent | $0.3 / $1.2 |
| 18 | Step 5 Preview | Other | 31.82% | $3.11 | Mini-SWE-agent | $1 / $2.7 |
| 19 | MiMo V2.6 Pro | Xiaomi | 31.31% | $0.50 | Mini-SWE-agent | $0.435 / $0.87 |
| 20 | Grok 4.7 | xAI | 28.79% | $18 | Mini-SWE-agent | $2 / $6 |
| 21 | Mistral Large 4 | Mistral | 27.78% | $10 | Mini-SWE-agent | $1.36 / $4.18 |
| 22 | MiniMax-M3.1-Flash | MiniMax | 25.25% | $0.73 | Mini-SWE-agent | $0.15 / $0.6 |
| 23 | Muse Spark 1.3 Max | Meta | 24.75% | $6.65 | Mini-SWE-agent | $1.25 / $4.25 |
| 24 | MiMo V2.6 Flash | Xiaomi | 24.24% | $0.22 | Mini-SWE-agent | $0.14 / $0.28 |
| 25 | Claude Opus 4.8 | Anthropic | 23.23% | $17 | Mini-SWE-agent | $5 / $25 |
Source: Vals AI leaderboard ↗ · Terminal-Bench project ↗
Best-ranked Anthropic model here: Claude Opus 5.5 (Anthropic). Try prompts for Claude →
Prompts for this kind of work
Copy-ready prompts from our library that match what Terminal-Bench 4.0 tests. Run them on your own material to compare models.
Michael | AI + Agents @Michael_Fenech_Report or remove