Open-weight & owned stack · open-lab-releases
DeepSeek released its V4 vision model and associated open-weight artifacts.
A vLLM Recipes entry documents DeepSeek's experimental V4-Flash-Vision model with a 32-layer vision tower, 1M context window, and fused DSpark draft module.
Read the original at recipes.vllm.aiOpens the publisher's site in a new tabAlso covering this
@zizhpanx.com, Sep 1DeepSeek Open-Sources First V4 Vision Model: Benchmark Claims Need Independent Prooftechtimes.com, Sep 1
More in Open-weight & owned stack
Abliteration.ai released API-accessible GLM-derived models with safety refusals removed.Sep 1MBZUAI’s Institute of Foundation Models released six fully open K2 Horizon models with weights, code, data, and methods.Sep 3Moonshot AI released Kimi K3 open weights, a technical report, and supporting training infrastructure.Sep 2NVIDIA reportedly released Nemotron 3 Ultra, a 550-billion-parameter open-weight model.Sep 3Z.ai released GLM-5.3, an open-weight model for coding, cybersecurity, and long-horizon agents.Aug 31