SWE-bench Verified: scores, leaders and methodology
SWE-bench Verified leaderboard data from the index. Real GitHub issues resolved by an AI coding agent, verified by human reviewers. The top score is 81.7 and the average is 57.5.
What the data shows
GPT-5.2 Pro leads at 81.7. The gap to the average is 24.2 points, and the gap to the 10th-ranked model is 2.2 points.
The median score is 61.0. Models above that line are in the upper half of the index for this benchmark.
How to read the ranking
- Use the score to shortlist, not to declare a winner.
- Check price and context for the top candidates.
- Read the benchmark methodology before quoting a score.
- Pair this benchmark with a task-specific evaluation.
Top 20 on SWE-bench Verified
| # | Model | Provider | Score | Input / 1M | Context |
|---|---|---|---|---|---|
| 1 | GPT-5.2 Pro | OpenAI | 81.7 | $21.32 | 2,000,000 |
| 2 | Claude Opus 4.5 Pro | Anthropic | 81.1 | $20.98 | 200,000 |
| 3 | GPT-5.2 Max | OpenAI | 80.5 | $21.39 | 1,000,000 |
| 4 | Claude Sonnet 4.5 Max | Anthropic | 80.2 | $19.90 | 1,000,000 |
| 5 | GPT-5.2 | OpenAI | 80.2 | $19.71 | 2,000,000 |
| 6 | GPT-5.1 Pro | OpenAI | 80.1 | $19.77 | 200,000 |
| 7 | o4 Max | OpenAI | 79.8 | $19.51 | 2,000,000 |
| 8 | Claude Sonnet 4.5 Reasoning | Anthropic | 79.6 | $19.48 | 400,000 |
| 9 | Claude Opus 4.5 Max | Anthropic | 79.5 | $20.90 | 1,000,000 |
| 10 | Claude Opus 4.5 Reasoning | Anthropic | 79.5 | $20.91 | 2,000,000 |
| 11 | GPT-5.2 Vision | OpenAI | 78.8 | $19.03 | 1,000,000 |
| 12 | Gemini 3 Pro | Google DeepMind | 78.5 | $18.24 | 2,000,000 |
| 13 | o4 Pro | OpenAI | 78.5 | $19.37 | 1,000,000 |
| 14 | o4 Turbo | OpenAI | 78.4 | $19.34 | 2,000,000 |
| 15 | GPT-5.2 Reasoning | OpenAI | 78.2 | $20.98 | 1,000,000 |
| 16 | Grok 4 Reasoning | xAI | 78.0 | $19.67 | 2,000,000 |
| 17 | Grok 4 | xAI | 77.9 | $17.90 | 200,000 |
| 18 | Claude Opus 4.5 Vision | Anthropic | 77.7 | $17.70 | 200,000 |
| 19 | GPT-5.2 Turbo | OpenAI | 77.4 | $20.73 | 1,000,000 |
| 20 | Gemini 3 Pro Turbo | Google DeepMind | 77.4 | $19.62 | 400,000 |
Methodology
This story uses the AI & LLM Index dataset: 600 models, 40 providers and 12 benchmarks. Prices are listed per million tokens. Scores are modelled estimates on a 0–100 scale and are not official benchmark results. Figures are rounded for readability.
Related
FAQ
What does the SWE-bench Verified: scores, leaders and methodology data show?
SWE-bench Verified leaderboard data from the index. Real GitHub issues resolved by an AI coding agent, verified by human reviewers. The top score is 81.7 and the average is 57.5.
Are the numbers official?
No. Prices and scores are compiled for this demo index. Benchmark values are modelled estimates, not official results. Verify figures with the provider or benchmark owner before making decisions.
How should I use this ranking?
Use it to form a shortlist. Check the specific model pages, compare price and context, and test the shortlist on your own examples before committing.