✗ Codex ran out of room in the model's context window. Start a new thread or clear earlier history before retrying.

Codex prints this the moment a running session’s history no longer fits the model’s context window. The trigger is accumulation: turns, tool output, and file reads pile up over a long run until they stop fitting, and Codex hard-stops instead of continuing. The generic “one input is too big for a single turn” case lives on a separate page, input exceeds context window. This page covers the runtime string you hit after a long session.

Symptoms

• The exact string shows up mid-session, often hours in. One reporter ran /goal for several hours on CLI v0.133.0 with gpt-5.5 before it started firing repeatedly (#24247).
• Retrying the same prompt in the same thread throws it again, because the old history is still loaded.
• Some people hit it almost immediately on a fresh session (one question in on CLI v0.80.0, #9046), which points at config rather than accumulation.
• Older IDE builds print slightly different wording, “Start a new conversation…” on VS Code extension 0.4.49 (#7808), versus “Start a new thread…” in the current CLI.

Cause

The running thread’s tokens went past the model’s context window. Codex used to auto-compact (summarize older turns) before hitting the wall; a change reported around June 18 2026 made it hard-stop with this message instead (community thread). Auto-compaction can also just fail to fire during long autonomous runs like /goal (#24247). In one forum case the real cause was broken config: model_context_window and model_auto_compact_token_limit overrides set on a built-in model, where those overrides don’t apply (they only work with custom providers), so compaction never kicked in (auto-compaction thread).

Fix

1. Run /compact. It summarizes the visible conversation to free tokens and lets you keep the task’s context (docs). Best first move when you don’t want to lose your place.
2. Run /new to start a fresh task inside the same CLI session, dropping the old history (docs).
3. Run /clear to wipe the terminal and start a clean task (docs).
4. Load less next time. Fewer and smaller file reads, and break a multi-hour /goal into staged runs so history has less to accumulate.
5. If it fires seconds into a new session, audit your config and remove model_context_window / model_auto_compact_token_limit overrides unless you’re actually on a custom provider (auto-compaction thread).

Affected versions

Reported on Codex CLI from v0.80.0 (#9046) through v0.133.0 (#24247), and the VS Code extension 0.4.49 (#7808). No release marks it fixed; the hard-stop behavior showed up around mid-June 2026, so it’s not tied to one version.

Facts on this page verified July 15, 2026.