DeepSeek ships V4.1-Flash with native vision; builds landing pages for 1/40th of Claude's cost
togethercompute · x · 2026-09-12
DeepSeek officially introduced DeepSeek-V4.1-Flash, the smallest model in its new architecture family:
- Native visual understanding
- Designed for higher capability, faster inference and higher throughput, with scaling headroom for larger models in the family
Developer Nutlope benchmarked it against Claude Fable 5 on the same prompt for building a movie-theater landing page: Fable cost $1.21 while V4.1 Flash cost just $2.6 cents — over 40x cheaper at similar quality, which he called "insane at landing pages."
Related event: DeepSeek releases V4.1-Flash with native vision and 1M context(3 posts)→
More from coding & agent
- AI infra conversations shift from GPUs to CPUs as agents move bottlenecks to orchestration — AccBalanced · 2026-09-12
- Yutori's Navigator n2 runs browser agents at $1.46 per task on OSWorld 2.0 vs $13-$40+ for frontier models — DhruvBatra_ · 2026-09-12
- Sourcegraph's astra coding agent is 'insanely good' at fixing memory leaks — weswinder · 2026-09-12
- Using astra x ChatGPT to audit and cancel all your subscriptions — with prompt — yungcontent · 2026-09-12
- Vibe-coded parkour game built with GPT-6 draws real players back 20+ times for leaderboards — TAbrodi · 2026-09-12
- OpenRouter's automatic fallbacks can bite: same model, different provider behavior — use provider.only — Simon Willison · 2026-09-12