feat(aihubmix): add qwen3.7-plus/step-3.7-flash and fix hy3 reasoning control#3427
feat(aihubmix): add qwen3.7-plus/step-3.7-flash and fix hy3 reasoning control#3427orangeclk wants to merge 3 commits into
Conversation
aihubmix gateway rejects boolean `thinking` for Hunyuan (hy3) with 400: "cannot unmarshal bool into ... model.Thinking". The `thinking` field must be an object {type:enabled|disabled}, which a Dify yaml parameter rule cannot emit.
Switch hy3.yaml from `thinking`(boolean) + `thinking_budget`(int) to a single `reasoning_effort` string parameter (none/low/medium/high, default none). Verified on aihubmix: all four values return 200; none disables thinking (reasoning_tokens=0), low/medium/high enable it. hy3 + reasoning_effort:high + json_schema end-to-end returns 200 with schema-conformant JSON.
Per Tencent Hunyuan docs, hy3 reasoning_effort supports low/medium/high (default low); none is an aihubmix extension to turn thinking off. llm.py is unchanged.
dosubot
Bot
added
size:XS
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request updates the aihubmix plugin to version 0.0.38, adding support for the new models qwen3.7-plus and step-3.7-flash, and updating the hy3 model configuration to use reasoning_effort instead of thinking parameters. The review feedback highlights that standard parameters (temperature, top_p, and max_tokens) were accidentally removed from hy3.yaml and are also missing from the newly added qwen3.7-plus.yaml and step-3.7-flash.yaml configurations. These parameters should be restored or added to ensure users can configure them in the UI.
Important
The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.
1 participant
Summary
Two aihubmix model-catalog updates bundled together:
reasoning_effort. The aihubmix gateway rejects a booleanthinkingfield for Hunyuan (hy3) with HTTP 400.qwen3.7-plusandstep-3.7-flashmodels. Both are multi-modal (vision + video) chat models with tool-call / stream-tool-call / multi-tool-call support,response_format(text / json_object / json_schema) and ajson_schematemplate rule.qwen3.7-plusexposes anenable_thinkingboolean (default off);step-3.7-flashexposes areasoning_effortstring (low/medium/high, defaultmedium). Registered both in_position.yaml.Change Type
Screenshots / Videos
LLM Plugin Checklist
Areas affected by this change (check all that apply)
Version
versioninmanifest.yaml(not the one undermeta) —0.0.37→0.0.38dify_plugin>=0.3.0,<0.6.0is declared inpyproject.tomland locked inuv.lock(or kept inrequirements.txtfor legacy plugins withoutuv.lock) — no dependency changes were needed for this change;pyproject.toml/uv.lockare untouched. (aihubmix already declaresdify_plugin>=0.9.0from feat(aihubmix): align with Anthropic official plugin v0.3.24 #3422.)Testing