MLX Benchmark V2 - QA: leaderboard
MLX Benchmark V2 question-answering slice for Apple MLX framework APIs and workflows.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Not forecast. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 89.35 | #90 |
| 2 | Gemini 3 Flash (Preview) | 84.4 | #82 |
| 3 | Qwen 3.6 Max Preview | 79.85 | #85 |
| 4 | Gemma 4 26B A4B (IT) | 76.39 | #297 |
| 5 | GPT-5.4 Nano | 76.39 | #328 |
| 6 | DeepSeek V4 Flash | 75.93 | #137 |
| 7 | DeepSeek V4 Pro | 75.5 | #102 |
| 8 | Grok 4.1 Fast | 73.15 | #217 |
| 9 | Gemini 2.5 Flash Lite (Preview 09-2025) | 68.75 | #318 |
| 10 | Qwen 3.6 35B A3B | 53.24 | #226 |
| 11 | GPT-5 Nano | 44.44 | #425 |
| 12 | GLM-5.1 | 17.59 | #97 |
| 13 | Nemotron 3 Ultra | 13.66 | #235 |
| 14 | Kimi K2.5 | 0.69 | #144 |
| 15 | Kimi K2.6 | 0.47 | #104 |
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-qa · How It Works · Data refreshed daily, snapshot 2026-10-20.