Anthropic's S-1 says over 90% of its September code contributions were written by Claude

The filing states that "over 90% of the contributions to our codebase in September 2026 were written by Claude." It adds that average daily code contributed per engineer in the three months ended September 30, 2026 was over 45 times the average for the year ended December 31, 2024, and that engineers now "direct and review" Claude Code's work rather than write code line by line. It also reports that Claude's success rate on open-ended engineering problems reached about 90% in September 2026, up about 50 percentage points in nine months, on internal benchmarks. The passages we read do not define a contribution or describe those benchmarks.

The filing also gives cost figures. It says Claude Fable 5.1 cut the cost of typical workloads by about 25% against Fable 5, with savings of up to 45% for complex coding and highly agentic tasks, and that Opus 5.5 cut cost per task by about 40% against Opus 5, through a mix of lower prices and better efficiency. For usage, it says the top 5% of Claude Enterprise users who used Claude Code spent about 20 times what Enterprise users who did not use it spent, which it puts below $75 a month on average.

Why it matters: These are the company's own numbers about its own engineering, so they show what one agent-heavy team reports, not what a typical team will see. Two things follow for anyone running coding agents. If review is where engineers now spend their time, verification capacity is the limit to plan around, not generation. And the stated per-task price drops mean any loop budget set before September is worth recalculating.