Four Chinese labs ship frontier-grade open-weights coding models in 12-day window — Air Street Press
Z.ai's GLM-5.1, MiniMax M2.7, Moonshot's Kimi K2.6 and DeepSeek V4 all landed within a fortnight, hitting near-frontier benchmarks on agentic coding at under a third of Claude Opus 4.7's inference cost.
Positive-sum angle: Open weights at the frontier let any developer, anywhere, run capable coding agents on local hardware. Each release lifts the floor for everyone downstream (start-ups, schools, public sector) instead of concentrating gains inside two or three labs.
What's the impact: SEA founders can now run frontier-grade coding agents on local hardware at under a third of Claude Opus inference cost. Plan tooling roadmaps assuming commodity inference, not lab-tier API spend, and budget the savings into product velocity.