An open model matches the frontier on kernels and autonomously designs a chip

Claim

Moonshot published evaluations showing Kimi K3 trails Fable 5 and GPT-5.6 Sol on aggregate yet beats everything else tested, taking outright wins on four benchmarks while matching Fable 5 on GPU kernel optimization, building a from-scratch compiler, and autonomously designing a chip.

Reported metric: $0.94 per index task

Reporting date: 2026-07-17 (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

• Welcome to July 17, 2026 (theinnermostloop.substack.com)