A surprise release puts frontier-level coding in reach of small teams — here's what changes for builders.
Updated Nov 6, 2026 · 3 min read
The 10-second version
The release landed without a press cycle, and within hours developers were posting side-by-side comparisons.
Early tests suggest it handles multi-file refactors noticeably better than previous open models, though long-context reasoning still trails the top closed systems.
Cheaper, self-hostable coding models mean small agencies can ship AI features without per-token bills eating their margin.