According to CoinMeta, a new artificial intelligence programming ranking list, FrontierCode, has been launched to assess whether AI code meets the merger standards set by open-source project maintainers. Claude Opus ranked first with a comprehensive score of 52.2%, but only 58% of the attempts passed the strict checks, indicating that over 40% did not meet the criteria. The evaluation was developed by Cognition in collaboration with 36 open-source project maintainers, who selected 100 private tasks. The models were required to run these tasks three times within the same agent framework, and none of the models successfully completed all 9 tasks. GPT-6.1 and SOL scored 49.1%, which is only 3.1 percentage points lower than Opus's score of 55.5%. However, the cost per task for GPT-6.1 and SOL was 0.81 dollars, while Opus incurred a cost of 10.15 dollars per task, representing a difference of more than 12 times. Xiaomi's Mimo-V2.6-Pro scored 45% with a cost of only 0.18 dollars per task. The 52.2% figure represents the comprehensive score, not the actual PR merger rate. The evaluation utilized automated checks and GPT-6 and Luna model reviews, rather than relying on maintainers to review each item individually.