DeepSeek V4.1 Flash: 552B MoE that outruns V4 Pro at flash prices
DeepSeek's Causal Encoder-Decoder flagship — 552B backbone with only 8B active on input and 16B on output, native multimodal, 1M context — lands on AWS SageMaker JumpStart.
Why it matters
First open model on the CED architecture: KV cache roughly a quarter of V4 Flash, so long-horizon agents get flagship quality at flash cost; DeepSeek reports it beats V4 Pro on every agentic row.
What you could build with it
An indie developer could build a whole-codebase refactoring agent on V4.1 Flash: feed an entire repository into the 1M-token window, let the model plan multi-file edits and execute them through tools, at input prices far below flagship closed models.
Does it hold up?
Usable in production now: one-click SageMaker JumpStart deploys, vLLM support for self-hosting, and 295k+ Hugging Face downloads in a week; caveats are vendor-run benchmark claims and Artificial Analysis flagging the model as very verbose, which inflates output cost.
Built with DeepSeek V4.1 Flash
- AWS adds DeepSeek V4-Pro, V4-Flash and V4.1-Flash to SageMaker JumpStartarticle · AWS announcement Nov 11: all three MIT-licensed MoE models deployable from JumpStart with one-click setup; V4.1-Flash pitched as the agentic option for coding and vision.
- Together AI hits 1205 tokens/sec on DeepSeek V4.1 Flasharticle · Together AI reports its Together Compiler reaching 1205 TPS on DeepSeek V4.1 Flash, evidence of production-grade serving speed.
- Unsloth ships high-accuracy GGUF quants of DeepSeek-V4.1-Flash-1017article · Unsloth releases Dynamic v3.1 GGUF quantizations of the 1017 checkpoint on Nov 11-12, runnable in llama.cpp and Ollama on consumer hardware.
Learn more
First spotted on hf: source.
More AI models
VoiceStudio: open-source, fully-local ElevenLabs alternative for voice cloning, dubbing, and audiobooksdebpalash/VoiceStudio is a desktop app that runs voice cloning, voice design, video dubbing, dictation, transcription…models · JEV 0.75Edge0-145B: open 145B MoE that runs on a single consumer GPUEdge0 open-sourced 145B and 120B latent-MoE models derived from Qwen3.8-Flash-Next, plus an inference framework that…models · JEV 0.74Strata: run a 125B model on a gaming PCFree, MIT-licensed local inference engine that runs Qwen3.8-Flash-Next (125B MoE, about 6B active per token) on…models · JEV 0.73Google AI Edge Gallery: offline speech recognition and synthesis on-deviceGoogle added fully offline speech recognition and text-to-speech to AI Edge Gallery, with ML Kit APIs on Android and…models · JEV 0.73Soup: fine-tune an 8B LLM on a 4 GB laptop GPU from one YAML fileAn open-source CLI that fine-tunes LLMs from a single YAML file, with layer streaming that trains an 8B model on a 4…models · JEV 0.73Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.