Skip to content

Add gpt-5.5 catalog support (OpenAI in-house model, released 2026-04-23) #70854

@mjt0926

Description

@mjt0926

Summary

OpenAI released GPT-5.5 on 2026-04-23, available through the Codex OAuth flow for ChatGPT Plus/Pro/Business/Enterprise subscribers. As of 2026-04-24, OpenClaw v2026.4.22 (bundled @mariozechner/pi-ai 0.69.0) still does not include gpt-5.5 in its model catalog (models.generated.d.ts), which causes any openclaw.json that lists openai-codex/gpt-5.5 under agents.defaults.models to fail at gateway warmup with:

startup model warmup failed for openai-codex/gpt-5.5: Error: Unknown model: openai-codex/gpt-5.5

Repro

  1. Add openai-codex/gpt-5.5 to agents.defaults.models in openclaw.json
  2. Set agents.defaults.model.primary = openai-codex/gpt-5.5
  3. Restart the gateway container
  4. Observe the warmup error above — sessions cannot resolve the model

Proposal

Because OpenClaw is now an OpenAI-sponsored foundation project (Peter Steinberger joined OpenAI 2026-02-15), it seems reasonable for the OpenClaw distribution to carry an in-house catalog override for first-party OpenAI models rather than waiting for the independent pi-ai upstream (maintained by @badlogic) to ship a new catalog release every time OpenAI announces a Codex model.

Concretely, a small provider-side registry in extensions/openai/openai-codex-catalog.ts (or a sibling file) for OpenAI's own model identifiers would:

  • Remove the week-plus delay between OpenAI Codex release and OpenClaw supporting it
  • Keep pi-ai as the canonical catalog for everything else
  • Signal to users that OpenAI-branded models are first-class citizens in OpenClaw

Why this matters

  • Terminal-Bench 2.0: gpt-5.5 = 82.7% vs gpt-5.4 = 75.1% (+7.6pt)
  • SWE-Bench Pro: gpt-5.5 = 58.6%
  • 400K context, improved token efficiency vs 5.4 (per OpenAI announcement)

For OpenClaw users running Codex agents in production, the inability to route to gpt-5.5 is a real gap.

Environment

  • OpenClaw: ghcr.io/openclaw/openclaw:latest = v2026.4.22
  • @mariozechner/pi-ai: 0.69.0
  • OS: Ubuntu (docker)
  • Auth: openai-codex:default (oauth, ChatGPT Plus/Pro plan with gpt-5.5 entitlement confirmed)

Labels (suggested)

`enhancement`, `model-catalog`, `openai-models`


Happy to submit a PR if maintainers can point to the preferred integration pattern (e.g. catalog override vs. upstream pi-ai contribution).

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions