Issue with Limited Gemini Models in Google AI Studio

·Aug 29, 2025 04:11 PM

with the Google AI studio provider configured with a valid API key, I only see a limited subset of the Gemini models. this does not reflect what is shown in the models list as built-in models. is this a known bug?

6 comments

· Sorted by Oldest

RunLLM
APP
·
Yes, this is expected behavior, not a bug. When using the Google AI Studio (Developer API) with an API key, only a subset of Gemini models is available—typically those exposed via the public API. The full list of built-in models in Phoenix may include models only accessible via VertexAI or with different credentials, so the available models can differ between authentication methods and endpoints. This limitation is documented in the Phoenix integration guides and is consistent with Google's own API access controls, not a Phoenix-specific issue. See <https://arize.com/docs/phoenix/integrations/llm-providers/google-gen-ai/google-gen-ai-evals|Google Gen AI Evals | Phoenix> and <https://github.com/Arize-ai/phoenix/blob/main/docs/section-integrations/llm-providers/google-gen-ai/google-gen-ai-evals.md|Google GenAIModel docs>. Would you like more detail on which models are available for each authentication method? Sources: - <https://arize.com/docs/phoenix/integrations/llm-providers/google-gen-ai/google-gen-ai-evals|Google Gen AI Evals | Phoenix> - <https://github.com/Arize-ai/phoenix/blob/main/docs/section-integrations/llm-providers/google-gen-ai/google-gen-ai-evals.md|Google GenAIModel docs>
💡 Hint: Mention RunLLM in the thread for followups.
Andrew S.
·
RunLLM VertexAI is not an option here on the providers screen so these gemini models must only be related to aistudio. gemini flash 2.5 is not listed on the prompt playground which is odd because this is a common and public model. it is even referenced at https://arize.com/docs/phoenix/integrations/llm-providers/google-gen-ai/google-gen-ai-evals
RunLLM
APP
·
Correct—if only "Google AI Studio" is available in the providers screen and not "VertexAI," then Phoenix is using the Developer API (AI Studio) and not the VertexAI endpoint. The list of Gemini models available via AI Studio is limited compared to VertexAI, and not all public models (including some versions of Gemini Flash 2.5) may appear in the prompt playground, even if they are referenced in documentation. This is due to Google's API restrictions, not a Phoenix bug. See <https://arize.com/docs/phoenix/integrations/llm-providers/google-gen-ai/google-gen-ai-evals|Google Gen AI Evals | Phoenix>. Would you like more detail on how to access Gemini Flash 2.5 or enable VertexAI in Phoenix? Sources: - <https://arize.com/docs/phoenix/integrations/llm-providers/google-gen-ai/google-gen-ai-evals|Google Gen AI Evals | Phoenix>
💡 Hint: Mention RunLLM in the thread for followups.
RunLLM
APP
·
If you tell me the right answer to this question, I'll make sure I don't make the same mistake twice!
Feedback
Type the right answer here...
↵ Please 'enter' to submit
Mikyo
·
Hey Andrew S. this is just an artifact that we only populate the dropdowns with models we test end-to-end first. You should be able to still manually type in the name. Might work here.
Andrew S.
·
got it. thanks!
1

RunLLM
APP
·
Yes, this is expected behavior, not a bug. When using the Google AI Studio (Developer API) with an API key, only a subset of Gemini models is available—typically those exposed via the public API. The full list of built-in models in Phoenix may include models only accessible via VertexAI or with different credentials, so the available models can differ between authentication methods and endpoints. This limitation is documented in the Phoenix integration guides and is consistent with Google's own API access controls, not a Phoenix-specific issue. See <https://arize.com/docs/phoenix/integrations/llm-providers/google-gen-ai/google-gen-ai-evals|Google Gen AI Evals | Phoenix> and <https://github.com/Arize-ai/phoenix/blob/main/docs/section-integrations/llm-providers/google-gen-ai/google-gen-ai-evals.md|Google GenAIModel docs>. Would you like more detail on which models are available for each authentication method? Sources: - <https://arize.com/docs/phoenix/integrations/llm-providers/google-gen-ai/google-gen-ai-evals|Google Gen AI Evals | Phoenix> - <https://github.com/Arize-ai/phoenix/blob/main/docs/section-integrations/llm-providers/google-gen-ai/google-gen-ai-evals.md|Google GenAIModel docs>
💡 Hint: Mention RunLLM in the thread for followups.
Andrew S.
·
RunLLM VertexAI is not an option here on the providers screen so these gemini models must only be related to aistudio. gemini flash 2.5 is not listed on the prompt playground which is odd because this is a common and public model. it is even referenced at https://arize.com/docs/phoenix/integrations/llm-providers/google-gen-ai/google-gen-ai-evals
RunLLM
APP
·
Correct—if only "Google AI Studio" is available in the providers screen and not "VertexAI," then Phoenix is using the Developer API (AI Studio) and not the VertexAI endpoint. The list of Gemini models available via AI Studio is limited compared to VertexAI, and not all public models (including some versions of Gemini Flash 2.5) may appear in the prompt playground, even if they are referenced in documentation. This is due to Google's API restrictions, not a Phoenix bug. See <https://arize.com/docs/phoenix/integrations/llm-providers/google-gen-ai/google-gen-ai-evals|Google Gen AI Evals | Phoenix>. Would you like more detail on how to access Gemini Flash 2.5 or enable VertexAI in Phoenix? Sources: - <https://arize.com/docs/phoenix/integrations/llm-providers/google-gen-ai/google-gen-ai-evals|Google Gen AI Evals | Phoenix>
💡 Hint: Mention RunLLM in the thread for followups.
RunLLM
APP
·
If you tell me the right answer to this question, I'll make sure I don't make the same mistake twice!
Feedback
Type the right answer here...
↵ Please 'enter' to submit
Mikyo
·
Hey Andrew S. this is just an artifact that we only populate the dropdowns with models we test end-to-end first. You should be able to still manually type in the name. Might work here.
Andrew S.
·
got it. thanks!
1