Anthropic: 80% of Merged Code Is Claude-Written
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic said 80%+ of code merged into its codebase is now authored by Claude, with engineers shipping 8x more code per quarter than during 2021-2025.
- Anthropic reported its Mythos Preview model improved on human researchers' next-step decisions 64% of the time (up from 22% in 2024) and achieved ~52x speedup on AI training code optimization in April, compared to Claude Opus 4's ~3x speedup in May 2024.
- Anthropic called for a global pause in AI development, saying recursive self-improvement is "approaching faster than we expected," while caveating that Claude's "research judgment" remains unclear.
- MakeUseOf reported Anthropic made Claude worse for a month before being caught — a degradation counterpoint largely overshadowed by the bullish capability narrative in wider coverage.
Why it matters: Anthropic is simultaneously building toward and warning about recursive self-improvement, a rare posture for a frontier lab. The 8x code throughput and 52x training-speedup numbers are concrete productivity gains, while the public call for a global pause puts competitive pressure on rivals who decline to slow down. Customers and competitors must now parse whether the cautionary framing is a regulatory hedge or a genuine constraint on deployment.
Ask SkimNews
