Alibaba's Qwen3.8-Max runs autonomous chip design for 60 hours, cutting power 59.5%
千问大模型 · wechat · 2026-09-22
Qwen announced two production-grade "chip-model co-evolution" experiments with Qwen3.8-Max. In chip design, given only a NoC module spec, the model autonomously ran 60+ hours with 10k+ tool calls through RTL, self-built verification, and physical implementation — cutting standard cells 29%, area 42%, and power 59.5%. In inference adaptation, it ported Qwen3.8-Flash-Next to the T-Head Zhenwu M890 in 56 hours with 40+ sub-agents, halving TTFT (−47%), cutting TPOT 60%, and boosting single-instance throughput 96%, driven by 8 self-written kernels and CUDA Graph capture, with bit-exact outputs. Humans still set goals, acceptance criteria, and sign-off.
More from coding & agent
- Pigment Agents now reads and posts in Slack, inheriting user permissions — mattturck · 2026-11-12
- Map of the agent economy: 50+ startups across payments, spend control, identity, memory — venturetwins · 2026-11-12
- OpenCode launches Browse, an on-device browser agent with local logins — Saul_Loveman · 2026-11-12
- VentureTwins maps the B2A economy: dozens of agent infra and app companies in first series installment — venturetwins · 2026-11-12
- Google releases 76-page 'Agents Companion' whitepaper on agent engineering — mdancho84 · 2026-11-12
- Spinning up containers for 20-line agent tools is backwards: Safescript compiles provable, 50ms-deploy apps — uriwa · 2026-11-12