LLM Stats (DeepSWE): leaderboard
LLM Stats mirror of DeepSWE — agentic software engineering where models autonomously resolve real-world coding issues end to end via the mini-swe-agent harness in isolated containers; resolved-task rate.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Around January 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol | 72.7 | #19 |
| 2 | GPT-5.6 Terra | 69.6 | #41 |
| 3 | Kimi K3 | 67.5 | #30 |
| 4 | GPT-5.6 Luna | 67.2 | #68 |
| 5 | Hy4 preview | 64.3 | #38 |
| 6 | DeepSeek V4 Pro (0813) | 62.7 | #53 |
| 7 | DeepSeek-V4-Flash-Vision-Exp | 59.3 | #76 |
| 8 | DeepSeek V4 Flash (0731) | 54.4 | #78 |
| 9 | Grok 4.5 | 53 | #48 |
| 10 | GLM-5.2 | 46.2 | #73 |
| 11 | Seed 2.1 Pro | 32.7 | #67 |
| 12 | Hy3 | 28 | #93 |
| 13 | Seed 2.1 Turbo | 23 | #86 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=llm-stats-deepswe · How It Works · Data refreshed daily, snapshot 2026-10-20.