Best LLM for long autonomous tasks
- Claude Opus 4.6 leads
- 78.9%
- models ranked
- 21
- lead over #2
- +1.8 pts
METR time horizons · read October 21
As of October 21, 2026, Anthropic Claude Opus 4.6 leads 21 models on METR time horizons with 78.9%, ahead of Google Gemini 3.1 Pro Preview (77.0%) and OpenAI GPT-5.2 (75.3%).
This list narrows the best LLM for agents ranking.
| Rank | Model | METR time horizons | Coding evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.6 Anthropic | 78.9% | 66.9 | $25.00 | |
| 2 | Gemini 3.1 Pro Preview Google | 77.0% | 35.3 | $12.00 | |
| 3 | GPT-5.2 OpenAI | 75.3% | 58.9 | $14.00 | |
| 4 | Claude Opus 4.5 Anthropic | 75.0% | 66.8 | $25.00 | |
| 5 | GPT-5.3-Codex OpenAI | 74.5% | 29.5 | $14.00 | |
| 6 | GPT-5.4 OpenAI | 74.3% | 37.7 | $15.00 | |
| 7 | GPT-5.1-Codex-Max OpenAI | 70.8% | Not benchmarked | $10.00 | |
| 8 | GPT-5 OpenAI | 69.6% | 72.2 | $10.00 | |
| 9 | Claude Sonnet 4.5 Anthropic | 67.4% | 59.1 | $15.00 | |
| 10 | Claude Opus 4.1 Anthropic | 66.8% | 27.2 | $75.00 | |
| 11 | o3 OpenAI | 65.4% | 70.7 | $8.00 | |
| 12 | o4 Mini OpenAI | 63.9% | 68.6 | $4.40 | |
| 13 | Kimi K2 Thinking Moonshot AI | 59.2% | 63.4 | $2.50 | |
| 14 | gpt-oss-120b OpenAI | 56.6% | 34.5 | $0.60 | |
| 15 | Gemini 2.5 Pro Preview 06-05 Google | 55.4% | 83.1 | $10.00 | |
| 16 | R1 0528 DeepSeek | 53.8% | 71.4 | $2.15 | |
| 17 | o1 OpenAI | 51.1% | 61.7 | $60.00 | |
| 18 | DeepSeek V3 0324 DeepSeek | 49.6% | 42.0 | $1.14 | |
| 19 | GPT-4o (2024-05-13) OpenAI | 40.8% | Not benchmarked | $15.00 | |
| 20 | Qwen2.5 72B Instruct Qwen | 35.8% | Not benchmarked | $0.40 |
How this list is built
Generated from the dataset on every refresh. Nobody picks the order by hand.
Models by their METR time horizons result, as the board publishes it, best first. METR measures how long a software task, timed on human experts, a model can finish with 50% reliability. Epoch AI publishes the score it derives from METR results.
- OpenRouter, fetched
- Epoch AI benchmarking hub, fetched