Open, specialized, and frontier models contest price-performance
OpenAI’s GPT-6.2 Astra reportedly scores 62 on Artificial Analysis versus Claude Fable 5.2’s 65, at reported costs of $4 versus $26.54 per task; Sol is said to approach Astra on FrontierCode 2.0 at about one-fifth the testing cost. Microsoft’s MAI models add lower-cost coding and cybersecurity claims, while open and specialized contenders continue to make benchmark and inference-efficiency claims. Independent, workload-matched replication remains the key test; benchmark configuration and introductory pricing can materially change the comparison.
Sources (34)
Updated Nov 18, 2026