Skip to content

Release 0.4.5 - #183

Merged
hiyouga merged 3 commits into
mainfrom
dev
Aug 20, 2026
Merged

Release 0.4.5#183
hiyouga merged 3 commits into
mainfrom
dev

Conversation

@hiyouga

@hiyouga hiyouga commented Aug 20, 2026

Copy link
Copy Markdown
Contributor

0.4.5 carries three PRs' worth of work; changelog/0.4.5/README.md lists all ten entries.

Highlights

  • Streaming clients no longer end a stream over output they do not recognize (#181). An unknown event is marked unused and an unknown Gemini content part is skipped; AGENTHUB_DEBUG set to anything other than 0/false/no/off restores the guards. This closes the gateway-heartbeat class of failure that 0.4.3 only half-closed: each family had learned its own spelling, so a relay's Anthropic-style ping still killed an OpenAI Responses stream.
  • AutoLLMClient.list_models() (#181) returns the model ids an endpoint serves, filtered to what the routed client can be used for. The playground lists them straight into its model dropdown.
  • default_headers / defaultHeaders (#181) on every client, for endpoints that reject a client they do not recognize.
  • GLM-5.3, claude-opus-5, and gpt-5.6-sol join the registry, and the OpenRouter entries move to the openai-responses client (#180).
  • UnsupportedOperationError joins the error classes, reporting a capability the routed client does not have.

Behavior changes worth reading

  • An unknown stream event is now skipped rather than raised unless AGENTHUB_DEBUG is set.
  • openai_responses and minimax_m3 now flush buffered text at the end of a message, like gpt5_6, so a turn of [text, tool_call] replays as [function_call, message].
  • The playground defaults AGENTHUB_DEBUG to 1.

Versions are already at 0.4.5 in src_ts/package.json and src_py/pyproject.toml.

🤖 Generated with Claude Code

https://claude.ai/code/session_013STCF9fADwqT8jJH9C3g5h

hiyouga and others added 3 commits August 20, 2026 16:05
…sts (#180)

GLM-5.3 launched, so its docs snapshot, registry pricing, and env-gated E2E model
move off the pre-launch footing they shipped on in 0.4.2. The docs snapshot is
re-taken as raw markdown from docs.z.ai and grows from 2 pages to 7; no client
change was needed, since the launched contract matches what the client was
written against.

Also in this change:

- claude-opus-5 joins the registry for the official API and OpenRouter.
- The official GPT-5.6 entry is named gpt-5.6-sol instead of the bare alias, and
  the three stale OpenRouter GPT-5.6 prices are refreshed from the live API.
- The 17 OpenRouter entries on the generic openai-chat client move to
  openai-responses, each verified against the live endpoint; SiliconFlow keeps
  openai-chat, the only protocol it serves.
- The playground model dropdown follows the E2E model list, and an option can
  declare its client type.
- The live E2E suites retry rate-limited responses; every other failure,
  assertion failures included, still fails on the first attempt.

The ant-messages Z.AI protocol mode stays on glm-5.2: that gateway defaults to
thinking disabled, which glm-5.3 rejects.

Closes #178
… and prepare 0.4.5 (#181)

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Copilot AI lite review requested due to automatic review settings August 20, 2026 15:59

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

@hiyouga
hiyouga merged commit 8a33b01 into main Aug 20, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants