Support for Gemini 3 Flash/Pro Preview and optimal configuration for reasoning models

[whosdum]

Hi everyone,

I am using the free version of Qodo (PR-Agent) and I would like to ask a few questions regarding the Gemini model support and configuration:

Are gemini-3-flash-preview and gemini-3-pro-preview already supported in PR-Agent? If so, what are the correct identifiers to use in the configuration?

Regarding the [config] section, I am currently using the following setup:

[config]
model="gemini/gemini-2.5-pro"
fallback_models=["gemini/gemini-2.5-flash"]
#model_reasoning="" 
#model_weak=""

My questions about the configuration:

If I decide to use reasoning models (like the new Gemini models or others), how should I ideally set up the model_reasoning and model_weak parameters?

How does PR-Agent decide which model to use when model_reasoning is filled? Does it automatically offload specific tasks to the reasoning model, or does it use it for every request?

In terms of performance/quality, is it better to keep everything on a single strong model (like Gemini 2.5 Pro), or is there a significant benefit to splitting tasks between model and model_reasoning?

Thank you for your help!

https://github.com/qodo-ai/pr-agent/blob/main/pr_agent/settings/configuration.toml/

[damnthonyy]

Hi @whosdum :) ,

About your question

---

1. Gemini 3 support

Yes. PR-Agent supports gemini-3-flash-preview and gemini-3-pro-preview. Use these identifiers in your config:

Provider	Flash	Pro
Google AI Studio	gemini/gemini-3-flash-preview	gemini/gemini-3-pro-preview
Vertex AI	vertex_ai/gemini-3-flash-preview	vertex_ai/gemini-3-pro-preview

---

2. model_reasoning and model_weak

• model_reasoning is used only for the self-reflection step in /improve (code suggestions). If you leave it unset, the main model is used for that step.
• model_weak is used for lighter tasks: /describe, /ask, line questions, and changelog. If unset, the main model is used.

Example setup:

[config]
model = "gemini/gemini-2.5-pro"
fallback_models = ["gemini/gemini-2.5-flash"]
model_reasoning = "gemini/gemini-3-pro-preview"   # optional: only for /improve self-reflection
model_weak = "gemini/gemini-2.5-flash"            # optional: describe, ask, changelog

i hope i was able to help you : )

[damnthonyy]

naorpeled i wasn't wrong, do we agree ?