Best LLM for fixing GitHub issues
- Claude Opus 5 leads
- 97.0%
- models ranked
- 76
- lead over #2
- +0.6 pts
SWE-bench Verified (Vals AI) · read October 21
# SWE-bench Verified (Vals AI), read October 21
leaders = {
"Claude Opus 5": 97.0%, "DeepSeek V4 Pro 0813": 96.4%, "GPT-5.6 Sol": 96.2%,
}
As of October 21, 2026, Anthropic Claude Opus 5 leads 76 models on SWE-bench Verified (Vals AI) with 97.0%, ahead of DeepSeek V4 Pro 0813 (96.4%) and OpenAI GPT-5.6 Sol (96.2%).
The full best LLM for coding ranking combines this benchmark with others.
| Rank | Model | SWE-bench Verified (Vals AI) | Coding evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|
| 1 | Claude Opus 5 Anthropic | 97.0% | 83.0 | $25.00 | |
| 2 | DeepSeek V4 Pro 0813 DeepSeek | 96.4% | 56.1 | $1.98 | |
| 3 | GPT-5.6 Sol OpenAI | 96.2% | 73.5 | $20.00 | |
| 4 | Grok 4.6 xAI | 95.6% | 65.7 | $6.00 | |
| 5 | GPT-5.6 Terra OpenAI | 95.4% | 63.9 | $12.00 | |
| 6 | GLM 5.3 Z.ai | 95.4% | 70.6 | $4.40 | |
| 7 | Claude Fable 5 Anthropic | 95.0% | 79.3 | $50.00 | |
| 8 | Kimi K3 Moonshot AI | 93.4% | 62.3 | $15.00 | |
| 9 | GPT-5.6 Luna OpenAI | 93.0% | 58.0 | $1.20 | |
| 10 | GLM 5.3 Flash Z.ai | 92.0% | 57.7 | $0.50 | |
| 11 | DeepSeek V4 Flash 0731 DeepSeek | 88.8% | 51.8 | $0.28 | |
| 12 | Claude Opus 4.8 Anthropic | 88.6% | 65.0 | $25.00 | |
| 13 | Muse Spark 1.2 Meta | 86.6% | 55.7 | $4.25 | |
| 14 | Grok 4.5 xAI | 86.6% | 58.6 | $6.00 | |
| 15 | Qwen3.8 27B Qwen | 86.0% | 74.2 | $2.55 | |
| 16 | Qwen3.8 Max (0902) Qwen | 85.6% | 62.4 | $6.00 | |
| 17 | GLM 5.2 Z.ai | 82.8% | 61.1 | $4.40 | |
| 18 | GPT-5.5 OpenAI | 82.6% | 71.3 | $30.00 | |
| 19 | T Inkling Small Thinking Machines | 82.2% | 27.3 | $1.20 | |
| 20 | Claude Opus 4.7 Anthropic | 82.0% | 67.5 | $25.00 |
How this list is built
Generated from the dataset on every refresh. Nobody picks the order by hand.
Models by their SWE-bench Verified (Vals AI) result, as the board publishes it, best first. Real GitHub issues fixed with one shared agent harness, run by Vals AI; current models that the swebench.com board has not caught up with. Share resolved.
- OpenRouter, fetched
- Vals AI boards, fetched