Clarify gpt-5.3-codex-spark context window metadata

Open 💬 0 comments Opened May 18, 2026 by Hor1zonZzz

Summary

Please clarify the authoritative context-window metadata for gpt-5.3-codex-spark.

The Codex model list exposes gpt-5.3-codex-spark, but its public context-window value is hard to verify. The public bundled model catalog appears to include gpt-5.3-codex metadata, but I could not find an explicit gpt-5.3-codex-spark entry.

Existing user reports also show Spark status output with an effective context window around 258K, while some Codex remote-model tests use 273K default and 400K max-style context-window values. This makes it unclear which value is intended for Spark.

Expected behavior

Please document or expose the intended values for Spark:

  • total context window
  • effective window shown by /status
  • whether the effective value reserves output or compaction budget
  • whether model_context_window can override it and what max value it clamps to
  • whether Spark should appear in the public bundled model catalog

Actual behavior

Spark is visible as a selectable Codex model, but its context-window semantics are unclear from public docs and repo metadata.

Related

  • #19868
  • #11641
  • #18572
  • #13271

View original on GitHub ↗