MiniMax partners with vLLM creators' Inferact for day-zero production inference of open models
woosuk_k · x · 2026-10-23
MiniMax and Inferact (founded by the creators of vLLM) announced a strategic engineering partnership covering two areas:
- Day-zero vLLM support: the teams optimize and validate model-specific serving paths before each MiniMax release, so new open weights serve at production quality the day they ship. Decoding optimizations and compatibility testing are upstreamed to vLLM.
- Production inference: the Inferact Platform has served MiniMax M3 and M3.1 Flash production traffic since launch, routing requests through the MiniMax platform alongside MiniMax's own capacity after meeting vendor verification requirements, with guaranteed SLAs for agentic workloads and auto-tuning against live traffic.
The partnership models a new "open weights + third-party inference stack" collaboration pattern for open-source model vendors.
More from Companies & People
- Shopify Post-Trains NVIDIA Nemotron to Beat Frontier Models at 90% Lower Cost — TimothyDuignan · 2026-10-23
- Ex-Meta product lead on life after Big Tech: parenting, games built by talking to ChatGPT — petergyang · 2026-10-23
- RadixArk caught paying KOLs to coordinate quote-posts of SGLang announcement — EugeneVinitsky · 2026-10-23
- Meta's Muse AI assistant shows 23% day-14 retention, but the comparison to ChatGPT is flawed — sujingshen · 2026-10-23
- Alvin Graylin joins UW as affiliate professor to teach AI and tech policy — GregCook2011 · 2026-10-23
- Nonprofit AI Research Org Basis Hiring Research Operations Manager — xuanalogue · 2026-10-23