Serve Qwen with vLLM: chat, tool calling and thinking
Pick a Qwen model that fits your GPU, start it with the right vLLM flags, then use thinking mode, tool calling and structured output from the OpenAI client.
Our guides and channel posts about OpenAI.
Number 6 this week: 431 points on Hacker News in the past 7 days and about 92 news articles a day.
Pick a Qwen model that fits your GPU, start it with the right vLLM flags, then use thinking mode, tool calling and structured output from the OpenAI client.
AI & technologyOpenAI and Google warn that AI search results and citations may be incomplete, outdated, or inaccurate, so verify the cited text, its dates, and authoritative sources.
AI & technologyAccording to OpenAI’s official pricing, GPT-6.1 Sol costs $2 per 1 million standard input tokens and $0.10 for cached input; cache hit rates can significantly change monthly costs for repeated prompts.
AI & technologyOpenAI announced on October 19, 2026, that it improved AI coding performance in low-resource programming languages such as COBOL, Fortran, and Visual Basic. The figures below are values reported by OpenAI from its own evaluation.
AI & technologyOpenAI’s official Deprecations document says the API shutdown date for several legacy models is October 23, 2026, including GPT-4 snapshots, GPT-3.5 Turbo, o1, o3-mini, and o4-mini. Requests using these IDs will fail afterward.
AI & technologyOpenAI broadly launched a new GPT-6-based ChatGPT experience on October 7. Paid plans default to GPT-6 Sol, while Free and Go default to GPT-6 Luna. Check the models used in ChatGPT, Codex, and Work separately.
AI & technologyOpenAI says it solved the Navier–Stokes problem, but independent review is still needed to verify the claim and assess the formalized Lean proofs.