Third-party evaluator Vals AI forecasts that Anthropic will reach full RSI in summer 2027. The background to the RSI measurement and the outlook for Anthropic’s timing are summarized.
The RSI Index published by Vals AI measures the ability of AI to develop the next generation of AI, calculating performance by assigning experiment planning, execution and evaluation to AI. The focus of this news is its forecast that Anthropic will reach full RSI in summer 2027.
In the official rankings, Anthropic’s Claude Opus 5.5 ranks first with an overall score of 0.37, and Claude models occupy the top 3 positions. Opus 5.5 is said to have previously surpassed human performance on a task involving improvements to the training methods of language models.
Krishnan said that the RSI Index results had improved at a nearly constant pace over the past year, and predicted that Anthropic would likely reach full RSI around July or August 2027. The timing is broadly unchanged from the forecast made 6 months earlier, but the evaluation covers only publicly available models, meaning that progress in actual development environments may be further advanced.
Anthropic co-founder Jack Clark has also brought forward his view, increasing the likelihood that RSI will be reached in 2027. Claude is reportedly writing most of the code in the company’s development environment. As with the third-party measurement results, 2027 is indicated as the target year.