NVIDIA releases DeepSeek V4.1 1017: 81 GB smaller, same performance
NVIDIA's DeepSeek‑V4.1‑Flash‑1017 model is 81 GB smaller with only a 0.7‑point drop in Terminal‑Bench and 0.26 in SciCode, keeping near‑original performance for production inference.
NVIDIA released the DeepSeek-V4.1-Flash-1017 version, whose size has been reduced by 81 GB with virtually no loss of quality.
In programming tests, the results dropped by only 0.7 points in Terminal-Bench and by 0.26 points in SciCode.
At the same time, the model's performance in programming and agent task benchmarks remained virtually at the previous level.
The optimized version is already ready for production use for inference.
Comments
Weekly digest
The best of vibe coding, AI agents and open source — once a week, no spam.
Related articles
All articlesDeepSeek V4.1 Flash 1017: 244 tps, $500/mo, Free API Search Till Oct 31
DeepSeek V4.1 Flash 1017 delivers 244 tokens per second, beating GPT‑6.1 Sol, priced at $500/month. Its web‑search API is free until Oct 31, then $0.75 per 1k requests.
NVIDIA Open-Sources LocateAnything-3B for Faster Visual Localization
LocateAnything-3B uses parallel decoding to predict bounding boxes at once, improving detection in dense scenes and supporting UI, OCR and documents.
DeepSeek V4.1 Flash Now Free on OpenCode – 200k Token Limit
DeepSeek V4.1 Flash is now available for free on OpenCode, but the free tier caps the context window at 200,000 tokens.