GPT-5.2-Codex Native Context Compaction: Auto-Summarizes at 400K Token Threshold for Multi-Hour Agent Sessions
OpenAI·high signal
GPT-5.2-Codex automatically summarizes prior conversation and code history into a compact representation when the session approaches the 400K token limit, maintaining coherent project state across multiple context windows without manual intervention. This makes long-running refactors and migrations — previously broken by context overflow — reliable for the first time. Benchmarks: SWE-Bench Pro 56.4%, Terminal-Bench 2.0 64.0% as of January 14, 2026.