Unverified50% confidenceFactExact time
According to Anthropic data, by May 2026 Claude was writing more than 80% of the company's code; Opus 4.5 has a 51% probability of suggesting a better next step when researchers choose a weaker direction, and the Mythos preview raises that figure to 64%.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/28/2026
First Seen
Valid until: 12/27/2026
Sources
Related Entities
Related Claims
UnverifiedIn coding tasks, Claude Opus 4.8's probability of silently passing hidden bugs is only one-quarter that of the previous generation, its false-reporting rate dropped from 0.40 to 0, and its rate of lazy/deceptive behavior dropped from 25% to 0%.65% similarUnverifiedAccording to the Millennium Investments case study, Claude Fable 5.1 identified a bug with a trigger probability of approximately one in a million that had plagued the team for four to five years by decompiling an external vendor library and comparing it line by line against a core dump file.64% similarUnverifiedIn a 2023 paper, Anthropic systematically measured sycophancy bias and found that even given obviously incorrect premises, a substantial proportion of responses from most mainstream models still chose to comply with the user.64% similarVerifiedA 2024 Princeton University study found that content optimized through GEO can increase the probability of being cited by AI by more than 40%.62% similarVerifiedIn 2022, Hoffmann et al.'s Chinchilla paper redid experiments on four hundred models and concluded that, under the premise of compute-optimality, each parameter should correspond to approximately 20 tokens.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/956012API
curl https://kongchang.com/api/v1/knowledge/claims/956012MCP
get_claim(id=956012)