Gemini 2.5 Pro Preview (source snapshot unspecified)
Google DeepMind
Scale's SWE-Bench Pro private-set leaderboard and paper report a generative entry labeled only Gemini 2.5 Pro Preview, and the leaderboard's public version field is empty. Although the paper says models used the latest versions as of September 18, 2025, it gives no Gemini API identifier, so Atlas cannot pin the result to the 03-25, 05-06, or 06-05 preview snapshot and retains it as a benchmark entry.
Benchmark scores
| Benchmark | Score |
|---|---|
| SWE-Bench Pro (Private Dataset) | 10.1 |