DeepSeek-V4.1-Flash lands on Nebius: 552B MoE activating 8B with 1M context
Arindam_1729 · x · 2026-09-18
- DeepSeek-V4.1-Flash is now live on Nebius Token Factory, with native image understanding and an architecture built for faster inference on long, input-heavy workflows.
- It's a 552B MoE that activates just 8B parameters for input and 16B for output, with a 1M token context window.
- Developer Arindam built a quick real-time webcam app that feeds multiple frames to the model for scene summarization; he reports it ran smoothly with surprisingly strong results for such a small setup.
Related event: DeepSeek Unveils V4.1-Flash, Smallest Model with Native Vision(2 posts)→
More from Models
- Self-proclaimed ChatGPT co-inventor launches Jev, claiming 20-200x speed at 40-400x lower cost — threepointone · 2026-09-18
- Data-first lab claims 100% synthetic training data can build a true cognitive core — sh_reya · 2026-09-18
- Benchmark Heaven aggregates 100 benchmarks and 800 models, adds first Benchmaxxing score — airesearch12 · 2026-09-18
- Legora's take: there is no best model — lawyers write evals and Legora BAR picks the winner — soleio · 2026-09-18
- OpenAI unveils misalignment disclosure framework, publishes six reports on observed cases — coherence · 2026-09-18
- Inside GPT-6 Astra: 100k+ GPUs at Stargate and a Bet on 3D Embodied AI — udmrzn · 2026-09-18