ChatGPT 0Auth broke after upgrading from 2026.4.12 to 2026.4.14

[shadow tapir]

ChatGPT 0Auth broke after upgrading from 2026.4.12 to 2026.4.14

[blazing tree]

Yeah, you’re probably not imagining it.

A couple of verified things from the docs:

• Long sessions absolutely burn more tokens because OpenClaw keeps sending system prompt, history, tool results, attachments, etc.
  Docs: https://docs.openclaw.ai/reference/token-use
• Old tool output can bloat Claude sessions, so pruning/compaction matters.
  Docs: https://docs.openclaw.ai/concepts/session-pruning
• If you mean Claude Code / Claude CLI auth through OpenClaw, Anthropic auth is supported, but billing behavior can still be less predictable than plain API-key usage.
  Docs: https://docs.openclaw.ai/providers/anthropic
  https://docs.openclaw.ai/concepts/oauth

A few quick things to try:

• /status or openclaw status --usage
• /compact
• /new to start a fresh session if the thread got huge

And can you paste:

openclaw --version

Also, are you using Anthropic API key or Claude CLI/subscription auth? That changes the answer a lot.

If you want real-world “is it just me?” anecdotes too, I’d also ask in #users-helping-users, since this forum is better for troubleshooting than crowd sampling.

Ah, that’s a much clearer symptom.

A few verified things to check first:

1. Make sure you’re still on the Codex OAuth route

   • ChatGPT/Codex sign-in in OpenClaw is openai-codex/*, not openai/*.
   • If the model got switched to openai/gpt-5.4, ChatGPT OAuth will look “broken”.
   • Docs: https://docs.openclaw.ai/providers/openai
     https://docs.openclaw.ai/help/faq

2. Run doctor, it has Codex OAuth-specific checks now

   openclaw --version
   openclaw models status
   openclaw doctor --deep
   

   Doctor explicitly checks for:

• stale models.providers.openai-codex overrides that can shadow the built-in OAuth route after upgrades
  • TLS/certificate issues that break the OpenAI OAuth auth endpoint
  • expired/reused refresh tokens
    Docs: https://docs.openclaw.ai/gateway/doctor

3. If doctor says refresh_token_reused or invalid_grant
   re-login:

   openclaw models auth login --provider openai-codex
   

My first suspicion for an upgrade break like this is old models.providers.openai-codex transport config shadowing the newer built-in path. That’s a documented gotcha.

If you paste the output of: