Astra is a quiet force. Artificial Analysis has updated their benchmark twice in 4 days to reflect its real strength

Astra is proving to be a significant force, leading Artificial Analysis to update its benchmark twice in four days to accurately reflect its capabilities. A similar situation occurred with Sol, which prompted Arena to restructure its coding benchmark, add a full-stack benchmark, and update its webdev benchmark to better reflect real-world performance after Astra's release. This suggests that OpenAI may not be fully optimizing for benchmarks.