MLX Benchmark V2 - Very Hard: leaderboard
MLX Benchmark V2 very-hard split measuring Apple MLX framework knowledge on the most difficult questions.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Not forecast. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 92 | #90 |
| 2 | GPT-5.4 Nano | 82 | #328 |
| 3 | Gemini 3 Flash (Preview) | 76 | #82 |
| 4 | Gemma 4 26B A4B (IT) | 66 | #297 |
| 5 | Grok 4.1 Fast | 60 | #217 |
| 6 | DeepSeek V4 Flash | 52 | #137 |
| 7 | Gemini 2.5 Flash Lite (Preview 09-2025) | 50 | #318 |
| 8 | Qwen 3.6 35B A3B | 36 | #226 |
| 9 | GPT-5 Nano | 34 | #425 |
| 10 | DeepSeek V4 Pro | 29.41 | #102 |
| 11 | GLM-5.1 | 2 | #97 |
| 12 | Kimi K2.5 | 0 | #144 |
| 13 | Kimi K2.6 | 0 | #104 |
| 14 | Nemotron 3 Ultra | 0 | #235 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mlx-benchmark-v2-very-hard · How It Works · Data refreshed daily, snapshot 2026-10-20.