Python SDK: no PyPI release since June 3 — bundled runtime (0.137.0a4) doesn't expose GPT-5.6 models

Open 💬 5 comments Opened Jul 11, 2026 by JarryGabriel
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What issue are you seeing?

The Python SDK on PyPI has been static for ~5 weeks while the CLI moved on: the last uploads are openai-codex 0.1.0b3 and openai-codex-cli-bin 0.137.0a4, both from June 3, while the CLI is at rust-v0.144.1 (July 9). On the bundled 0.137 runtime, the GPT-5.6 family (GA July 9) never appears, so SDK users can't reach the current models.

What steps can reproduce the bug?

pip install openai-codex==0.1.0b3
python -c "import asyncio; from openai_codex import AsyncCodex
async def m():
    async with AsyncCodex() as c:
        page = await c.models()
        print([m.id for m in page.data])
asyncio.run(m())"
# → ['gpt-5.5', 'gpt-5.4', 'gpt-5.4-mini']   (no gpt-5.6 family)

What is the expected behavior?

A new SDK beta pinning a runtime ≥ 0.144 so models() exposes the GPT-5.6 family. (Third-party reporting suggests 5.6 is only served to CLI ≥ 0.144, which would match what we observe.)

Additional information

sdk/python on main is actively developed (commits through July 6), just unreleased. Staying pinned this far back also re-exposes deserialization skew like #21871 (still open).

View original on GitHub ↗

5 Comments

github-actions[bot] contributor · 1 month ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #31835
  • #32412

Powered by Codex Action

ebarti · 1 month ago

+1 to this issue. Please don't make us have to own our own SDK versions here...

ebarti · 1 month ago

@etraut-openai Any chance you could take a look at this?

JarryGabriel · 1 month ago

@ebarti The workaround is three steps:

  1. Install an up-to-date binary outside PyPI

npm i -g @openai/codex # → codex-cli 0.144.3

  1. Point the SDK at it — it already supports this natively:

from openai_codex import Codex, CodexConfig
cx = Codex(CodexConfig(codex_bin="/path/to/the/codex/binary"))

  1. Patch the ReasoningEffort enum in site-packages/openai_codex/generated/v2_all.py, otherwise models() blows up with a ValidationError (the 0.144 binary returns max/ultra efforts the frozen schema doesn't know):

max = "max"
ultra = "ultra"
After that, cx.models() returns gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna.

All of this is temporary — it goes away as soon as OpenAI ships a new PyPI release.