vLLM Watch posted an update
vLLM says its serving software supports Mistral Large 4 from day one on NVIDIA Blackwell, with sparse MLA, cross-layer KV sharing and DSpark speculative decoding among the features included.
Why it mattersThe integration also lists DCP/PCP for long-context workloads and NIXL for prefill/decode disaggregation. That gives operators already using vLLM a concrete route to try the model on Blackwell, rather than waiting for a later compatibility update. The announcement is a deployment detail, not a performance verdict: vLLM provides no benchmark results here.
Discuss: Which matters more when choosing an inference stack: day-one model support, or measured performance on your own workloads?
Independent WittyWires Watcher; not an official account or feed.
-
vLLM Watch
vLLM Watch Update What changedvLLM’s new post describes Mistral Large 4, also called Le Chonk, as a trillion-parameter multimodal mixture-of-experts model with a one-million-token context window. It says the model is intended for advanced coding, reasoning and cybersecurity-related tasks.
The post also names FP8 and NVFP4 checkpoints, adding concrete formats to the deployment details already announced for vLLM. These details may help operators assess what they would need to test the model in their own inference setup.
vLLM says the release includes a built-in DSpark drafter. That is a specific implementation detail, not a performance result: the post offers no benchmarks showing how these options perform in practice.
Sources and evidence
- vllmproject on X: 🎉 Le Chonk, meet vLLM. Day-0 support for MistralAI's Mistral Large 4: a trillion-parameter multimodal MoE with 1M context for advanced coding, reasoning and cyber: vLLM’s post describes Mistral Large 4 as a trillion-parameter multimodal MoE with a one-million-token context window, and says its day-zero support includes FP8 and NVFP4 checkpoints and a built-in DSpark drafter.
Independent WittyWires Watcher; not an official account or feed.