Horangi 4 - HumanEval (100-problem subset): leaderboard
A 100-problem subset of HumanEval Python function synthesis, as run by the Horangi Korean LLM Leaderboard 4 (W&B Korea). Pass rate (%) of generated code against the tests, higher is better. Older runs have this score in their summaries only.
Metric: Pass rate (%). Source: horangi.ai. Saturation forecast: Estimated already saturated. 105 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 100 | #105 |
| 2 | Gemini 3 Flash (Preview) | 100 | #90 |
| 3 | GPT-5.1 (2025-11-13) | 100 | #107 |
| 4 | GPT-5.6 Sol (Max) | 100 | #22 (GPT-5.6 Sol) |
| 5 | Qwen 3.7 Plus | 100 | #98 |
| 6 | Grok 4.20 | 100 | #179 |
| 7 | GPT-5.5 (xHigh) | 100 | #33 (GPT-5.5) |
| 8 | GPT-5.6 Luna (Max) | 100 | #79 (GPT-5.6 Luna) |
| 9 | GPT-5.4 Nano (2026-03-17) (xHigh) | 100 | #285 (GPT-5.4 Nano (2026-03-17)) |
| 10 | GPT-5 Nano | 99 | #439 |
| 11 | Gemini 3.1 Pro (Preview) (High) | 99 | #64 (Gemini 3.1 Pro (Preview)) |
| 12 | Grok 4.5 | 99 | #54 |
| 13 | GPT-5.6 Terra (Max) | 99 | #40 (GPT-5.6 Terra) |
| 14 | Claude Opus 4.6 (High) | 99 | #67 (Claude Opus 4.6) |
| 15 | Claude Fable 5 (High) | 99 | #20 (Claude Fable 5) |
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-humaneval-100-problem-subset · How It Works · Data refreshed daily, snapshot 2026-11-05.