🤖 OpenAI Cuts GPT-5.6 Agent Costs

GPT-5.6 Luna scored 84.04% on BrowseComp, nearly matching GPT-5.5 at 84.36%. Its run cost $1.33 instead of $33.27, making the gap hard to ignore.

For agent work, OpenAI recommends Luna and Terra for data extraction and document search. GPT-5.6 Sol on low reasoning also beat GPT-5.5 on high reasoning in one agent test.

The bigger gain comes from the agent setup. GPT-5.6 can preserve reasoning between steps and compact long histories. On ARC-AGI-3, that lifted Sol from 13.3% to 38.3% while using about 6 times fewer output tokens. The model itself was unchanged.

📊@tech