76.8%
#1
of SWE-bench Verified issues resolved by Claude 4.5 Opus (high reasoning), February 2026; the tenth-placed model resolved 70.8%
What the chart shows
SWE-bench Verified gives a model a real codebase and a GitHub issue, and counts the share of human-checked issues it resolves with a working patch. As of February 2026, Claude 4.5 Opus (high reasoning) led at 76.8%, ahead of Gemini 3 Flash and MiniMax M2.5 (both 75.8%) and Claude Opus 4.6 (75.6%).
The top 10 models sit within 6 percentage points of each other, down to Kimi K2.5 at 70.8%. On the older SWE-bench Lite subset, where entries mix agent systems and models, the AI Index’s top 10 ranged from 44% to 60.3%.
Key numbers
| Measure | Value |
|---|---|
| Claude 4.5 Opus (high reasoning) | 76.8% |
| Gemini 3 Flash (high reasoning) | 75.8% |
| MiniMax M2.5 (high reasoning) | 75.8% |
| Claude Opus 4.6 | 75.6% |
| Kimi K2.5 (high reasoning), 10th | 70.8% |
| Best on SWE-bench Lite: ExpeRepair-v1.0 + Claude 4 Sonnet | 60.33% |
How to read it
- All Verified results use the same agent workflow (mini-SWE-agent-v2), so differences reflect the model rather than the system around it.
- The leaderboard lists GPT-5.2 Codex twice, as “GPT-5-2 Codex” and “GPT 5.2 Codex”, both at 72.8%; we show both as published.
- The source figure also shows the top 10 on SWE-bench Lite, a smaller subset; those rows are in the CSV.
Source: SWE-bench Leaderboard, 2026, via the AI Index Report 2026 (Stanford HAI), figure 2.5.1: SWE-bench: percent solved. Chart and text by AI Adoption.
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Discussion