Ultrafast unavailable with API-key auth: Codex catalog omits tier that works in Responses API #52296

What version of Codex CLI is running?

codex-cli 0.162.0 (latest stable GitHub release at the time of testing, October 8, 2026).

What subscription do you have?

OpenAI API-key authentication with API token billing. This is not ChatGPT subscription authentication.

Which model were you using?

gpt-6.1-sol; also reproduced the API/catalog mismatch for gpt-6-astra.

What platform is your computer?

Linux x86_64; Linux 7.0.0-38-generic x86_64 unknown.

What issue are you seeing?

Ultrafast is usable through the public Responses API with our API key, but the Codex model catalog fetched with the same key advertises only Fast (priority). Consequently Codex does not expose Ultrafast, and its request resolver filters out a manually configured service_tier = "ultrafast" when the catalog lacks that tier.

This appears to be a mismatch between API eligibility and Codex backend model metadata, rather than missing API account access or a disabled local feature flag.

Direct live tests on October 8, 2026:

The catalog entries for both models contain:

{
  "service_tiers": [
    {
      "id": "priority",
      "name": "Fast",
      "description": "2x speed, increased usage"
    }
  ],
  "default_service_tier": null
}

The local cache also reflects this response. There is no model_catalog_json override configured.

What steps can reproduce the bug?

1. Authenticate Codex 0.162.0 with an API key eligible for Ultrafast.
2. Select gpt-6.1-sol or gpt-6-astra; confirm the catalog only advertises Fast.
3. With the same key, issue this direct API request:

curl https://api.openai.com/v1/responses \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-6.1-sol",
    "service_tier": "ultrafast",
    "input": "Reply with OK.",
    "max_output_tokens": 32,
    "reasoning": {"effort": "low"},
    "store": false
  }'

4. Check that the response completes with service_tier: "ultrafast".
5. Fetch Codex metadata using that same key:

curl 'https://chatgpt.com/backend-api/codex/models?client_version=0.162.0' \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H 'User-Agent: codex_cli_rs/0.162.0' \
  -H 'originator: codex_cli_rs'

6. Inspect service_tiers for both models: ultrafast is absent. Repeat the direct inference test with gpt-6-astra to verify the same mismatch.

The probes above were actually executed successfully; the filtering behavior below was verified by source inspection, not by capturing an outbound Codex inference request.

What is the expected behavior?

Codex's API-key model catalog should advertise ultrafast for accounts/models that can successfully use it through the Responses API. Codex should expose the option and preserve the selected tier in inference requests.

Please reconcile the Codex catalog's tier eligibility with API eligibility. If catalog omission is intentional, provide an actionable explanation instead of making a supported API tier unavailable.

Additional information

Source trace for the installed 0.162.0 release, commit c1382380de69521303b416720a52f42d51af6248:

• API-key metadata discovery routes to the Codex backend, rather than the public /v1/models endpoint.
• The UI builds tier commands from catalog service_tiers.
• The request resolver filters out unsupported catalog tiers. Flex has an explicit catalog-independent exception; Ultrafast does not.

Official Ultrafast API documentation states Ultrafast is broadly available for GPT-6 Astra and GPT-6.1 Sol.

A potential local workaround is model_catalog_json with ultrafast added to these models, but that pins a local catalog snapshot. This workaround has not been applied or validated here.

Codex doctor report

codex doctor --json completed with overallStatus: "ok". Relevant results summarized below; unrelated local paths, thread inventory, and environment details are omitted:

{
  "codexVersion": "0.162.0",
  "overallStatus": "ok",
  "auth": {
    "stored auth mode": "api_key",
    "stored API key": "true",
    "stored ChatGPT tokens": "false"
  },
  "config": {
    "model": "gpt-6.1-sol",
    "model provider": "openai",
    "enabled relevant features": [
      "api_key_model_discovery",
      "fast_mode",
      "ultrafast_mode"
    ]
  },
  "websocket": {
    "handshake result": "HTTP 101 Switching Protocols"
  },
  "updates": {
    "latest version": "0.162.0",
    "latest version status": "current version is not older"
  }
}