NVIDIA ships an 81GB FP4 rewrite of DeepSeek-V4.1-Flash with near-identical coding scores
evisoft · x · 2026-11-01
NVIDIA released an NVFP4-quantized DeepSeek-V4.1-Flash-1017 (81GB) on Hugging Face under MIT license. Coding scores drop only 0.7 and 0.26 points on Terminal-Bench and SciCode, with near-identical coding and agent benchmark performance, ready for production inference.
Poster evisoft highlights the irony: export controls block Nvidia's newest chips from China, so Nvidia rewrote China's best open model into the new chips' native format — the hardware is blocked, but the weights still serve the Nvidia ecosystem.
More from Infra
- Two of the ten largest AI data centers are in Malaysia, run by DayOne for ByteDance — Jsevillamol · 2026-11-01
- Dev Rants: Users Blame AI for RAM Prices While Refusing to Pay for ChatGPT — flowersslop · 2026-11-01
- Analyst bets on GPU overbuild, pricing idealized compute at $38/hour — bookwormengr · 2026-11-01
- Commentary: AI doesn't repeal economics — data center spend must prove its value — DavidLinthicum · 2026-11-01
- Musk confirms Colossus 2 will hit ~1.2M Blackwell GPUs by end of 2025, 5x Colossus 1 — XFreeze · 2026-11-01
- Ex-Nvidia exec: Meta's 40% idle GPUs could flood market, NeoClouds in trouble — SumitGup · 2026-11-01