[Enhancement]: Support for Gemini 3.5 Flash Model (gemini-3.5-flash)

[marlonka]

What features would you like to see added?

Google has officially released Gemini 3.5 Flash (gemini-3.5-flash) at Google I/O 2026. This model is now generally available via Google Agent Platform (formerly Vertex AI), and the Gemini API.
https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-5-flash
https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash
https://x.com/GoogleAIStudio/status/2056797490553540756

I would like to see full native integration of Gemini 3.5 Flash added to LibreChat's Google/Vertex AI provider configurations. This includes Agent Platform (Vertex AI) agent support to upload files via UploadToProvider to really fast upload PDF documents via the endpoint to the gemini 3.5 flash model.

Key technical requirements for this integration based on the official migration guidelines:

1. Model ID Support: Add gemini-3.5-flash to the list of available Google models.
2. Context Window & Specs: Configure the model settings to support its 1M token context window and a massive 65k maximum output token limit. (Input token limit 1,048,576 Output token limit 65,536)
3. Thinking Effort Configuration: Support the new thinking_level parameter instead of the deprecated thinking_budget. The default thinking effort for Gemini 3.5 Flash should be set to medium (which balances latency and quality for code and agentic tasks), with configurable parameters for minimal, low, medium, and high.
4. Configuration Cleanup: Remove legacy parameters like temperature, top_p, and top_k for requests sent to this specific model ID, as Google no longer recommends them for the 3.5 Flash architecture.

More details

Gemini 3.5 Flash represents a massive paradigm shift for fast-tier models. According to Google's release data, it actually outperforms Gemini 3.1 Pro on major coding and agentic benchmarks (such as Terminal-Bench 2.1 and MCP Atlas) while operating at 4x the speed of comparable frontier models and half the cost ($1.50 / $9 per 1M tokens).

Bringing this model to LibreChat will allow users to run massive, complex multi-turn chats and reasoning workflows directly within the LibreChat interface with incredibly fast token-per-second execution. Additionally, because Gemini 3.5 Flash features automatic "thought preservation" across API turns, it will natively enhance LibreChat's capability for long-horizon tasks like iterative code refactoring and deep document analysis.

Which components are impacted by your request?

Endpoints

Pictures

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

[NN1985]

Removing parameter support is a major blow to devs, especially coders who enjoy t=0 work. This move is not about performance - it's about anti-distillation efforts. Shame on Google.