AA Omniscience - Software Engineering (SWE) - Java: leaderboard
Metric: Accuracy (%). Source: artificialanalysis.ai. 294 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) (High) | 55 |
| 2 | Claude Opus 4.5 (Thinking) | 53 |
| 3 | Kimi K3 (Max) | 52 |
| 4 | DeepSeek V4 Pro (0813) (Reasoning, Max Effort) | 49 |
| 5 | Kimi K3 (Low) | 48 |
| 6 | Claude Opus 4.5 (Non-reasoning) | 45 |
| 7 | Kimi K2.7 Code | 43 |
| 8 | GPT-5 Codex (High) | 42 |
| 9 | DeepSeek V4 Pro (Reasoning, Max Effort) | 39.39 |
| 10 | Grok 4 | 38 |
| 11 | Hy3 | 37 |
| 12 | GPT-5 (Low) | 37 |
| 13 | GPT-5 (High) | 36 |
| 14 | DeepSeek V4 Pro (Non-reasoning) | 36 |
| 15 | Kimi K2.5 (Thinking) | 35 |
Interactive version: theaggregate.ai/benchmark?slug=aa-omniscience-software-engineering-swe-java · How It Works · Data refreshed daily, snapshot 2026-09-05.