Horangi 4 - SWE-bench Verified (80-task subset): leaderboard
SWE-bench Verified, an 80-task subset evaluated with the official harness by the Horangi Korean LLM Leaderboard 4 (W&B Korea). Share of issues resolved (%), higher is better.
Metric: Resolved (%). Source: horangi.ai. Saturation forecast: Around February 2027. 105 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 73.75 | #90 |
| 2 | GPT-5.4 (2026-03-05) (xHigh) | 73.75 | #63 (GPT-5.4 (2026-03-05)) |
| 3 | GPT-5.5 (xHigh) | 70 | #33 (GPT-5.5) |
| 4 | Gemini 3 Pro (Preview) (High) | 68.75 | #71 (Gemini 3 Pro (Preview)) |
| 5 | Gemini 3.1 Pro (Preview) (High) | 67.5 | #64 (Gemini 3.1 Pro (Preview)) |
| 6 | GPT-5.4 Mini (2026-03-17) (xHigh) | 67.5 | #180 (GPT-5.4 Mini (2026-03-17)) |
| 7 | Claude Fable 5 (High) | 66.25 | #20 (Claude Fable 5) |
| 8 | Qwen 3.7 Plus | 66.22 | #98 |
| 9 | Grok 4.20 | 65.38 | #179 |
| 10 | GPT-5.6 Sol (Max) | 65 | #22 (GPT-5.6 Sol) |
| 11 | Gemini 3.5 Flash | 64.56 | #66 |
| 12 | Qwen 3.7 Max | 64.47 | #85 |
| 13 | Claude Opus 4.5 (20251101) (High) | 62.82 | #81 (Claude Opus 4.5 (20251101)) |
| 14 | Grok 4.5 | 59.49 | #54 |
| 15 | Claude Opus 4.6 (High) | 58.97 | #67 (Claude Opus 4.6) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=horangi-4-swe-bench-verified-80-task-subset · How It Works · Data refreshed daily, snapshot 2026-11-05.