Artificial Analysis has published the second version of its Coding Agent Index, which evaluates AI models based on their performance across 290 distinct software tasks. The updated index now features the FrontierCode v1.1 benchmark from Cognition, replacing the previous DeepSWE metric.
The latest rankings place Anthropic’s Claude Opus 5.5 at the top of the list with an overall score of 61.7. This assessment also incorporates revised grading criteria for the SWE-Atlas-QnA benchmark to better measure the ability of models to handle real-world code mergeability.
Market participants are currently tracking upcoming developments from major industry players, including Google and Meta. Future model releases or updated benchmark results from these competitors may shift current market expectations and influence pricing trends by the end of the month.
