Tool use across OpenAI models in a 616,300-response review

SemiAnalysis examined 616,300 OpenAI responses from its own internal usage. It reports that GPT 5.6 Terra averaged 1.94 client tool calls per response, versus 1.00 for Sol and 0.97 for Luna. Its early GPT 6 Astra usage fell between those rates, at 1.18 calls per response across 11,717 responses.