What Happened
OpenAI introduced GPT-5.1-Codex-Max as a new frontier model for agentic coding in Codex, highlighting long-running tasks and a compaction mechanism for sustaining work over very long sessions.
Why It Matters
This release illustrates the “agent reliability” direction: not just writing code, but persisting across hours-long workflows, iterating on tests, and maintaining coherence across repeated context windows.
Technical Details
- Long-horizon work: Designed for sustained, multi-step tasks
- Context strategy: Compaction to preserve key context while continuing work across long sessions
- Codex integration: Announced as available in Codex tooling surfaces