Model capability, evaluation integrity, and what inference actually costs.
OpenAI's GPT-6 Sol and GPT-6 Luna arrived on September 22 at half the promotional price of their GPT-5.6 namesakes. The rate card shows the Sol name now sits one rung lower than it did.
Anthropic released Claude Sonnet 5.5 on September 28 at Sonnet 5's prices. Its migration docs list five requests that now fail, and one of them collides with the model's new cyber fallback.
Claude 5 models think by default; the reasoning bills as output and counts against max_tokens. What flips when you swap the model string.
