An AI designed an open chip that runs small language models.
A developer published openTPU, an accelerator they say AI agents designed down to the wiring. On one Xilinx card it runs Qwen models at 21 to 86 tokens a second. The repository drew 240 stars since 24 September and the whole stack is open.
Why it mattersChip design is the one part of this boom that stays inside a few companies. If models can write hardware, small teams can try their own.
Read in the app — free, in 9 languages
Related stories
OpenAI's new endpoint picks from a fixed list in 150 milliseconds.2026-10-07Developers think a free mystery model belongs to Thinking Machines.
2026-10-06Mistral will give away its trillion-parameter model this month.
2026-10-06TikTok put a shopping chatbot inside the video you are watching.
2026-10-06Reflection will hand out a 501-billion-parameter model for free.
2026-10-06