Bug: gemini-3-flash-preview giving ChatModelError response

[sandijean90]

Describe the bug
When I switch my requirement agent to use gemini-3-flash-preview it gives me a chat model error. This is the only changed variable in the agent.

To Reproduce
Use gemini-3-flash-preview as the model with any requirement agent.

Expected behavior
Agent response from litellm, rather than chat model error

Set-up:

• Bee version v 0.1.74]
• Model provider gemini

Additional context

[dosubot]

This is a known issue that was fixed in Bee v0.1.74: Gemini models (including gemini-3-flash-preview) require prompt caching to be disabled, or else you’ll get a ChatModelError due to how RequirementAgent injects cache control metadata. The GeminiChatModel adapter now sets allow_prompt_caching=False by default, so if you’re using Bee v0.1.74+ and instantiating the model via ChatModel.from_name or GeminiChatModel, you should be covered automatically (source, source).

If you’re still seeing the error, double-check:

• Your virtual environment is fully updated and not using stale dependencies (source).
• GEMINI_API_KEY and GEMINI_CHAT_MODEL are set and valid in your environment (source).
• You’re not overriding allow_prompt_caching anywhere in your code.

If you can share the exact error message or stack trace, I can help pinpoint if there’s a new edge case or configuration issue.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

  

[Tomas2D]

I was able to reproduce. The ChatModel request hangs when using LiteLLM 1.80.10. Fortunately, after updating to LiteLLM 1.80.11, it started working. I will enforce the newer version in the upcoming release. Later today or tomorrow.

[Tomas2D]

Fixed and released in v0.1.75