NVIDIA DeepSeek-V4.1-Flash-1017-NVFP4 Card - MMMU-Pro: leaderboard
MMMU-Pro multimodal college-level questions; repeat count not stated. NVIDIA's DeepSeek-V4.1-Flash-1017-NVFP4 model card (Hugging Face, 2026-10-30): NVIDIA ran the source DeepSeek-V4.1-Flash-1017 checkpoint (MXFP4 routed experts) and its own NVFP4 quantization with vLLM at reasoning_effort=max, temperature 1.0 and top_p 1.0, to show the quantization's accuracy cost. Accuracy (%), higher is better.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around February 2027. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4.1 Flash (1017) (Max) | 78.38 | #27 (DeepSeek V4.1 Flash (1017)) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=nvidia-deepseek-v4-1-flash-1017-nvfp4-card-mmmu-pro · How It Works · Data refreshed daily, snapshot 2026-11-05.