Vercel Integrates Gemini 3.5 Flash for Parallel Agentic Workflows

May 19, 2026

Vercel, a frontend cloud platform and creator of the AI SDK, integrated Google's Gemini 3.5 Flash into its AI Gateway (a unified API for model management). Developers can now call the model using the google/gemini-3.5-flash identifier without managing separate Google Cloud or Vertex AI credentials.

Model identifier: google/gemini-3.5-flash
Thinking levels: low, medium, and high
Unsupported parameters: temperature, topP, topK, and thinking_budget
Pricing: No markup over Google base cost
Availability: Vercel AI Gateway and AI SDK

This integration mirrors the Google Gemini 3.5 Flash launch and reflects a shift toward reasoning-centric controls. By removing probabilistic parameters like temperature and top-P, the model focuses on thinking budgets to improve coherence. This aligns with data showing that agentic workloads drive most production traffic on Vercel's infrastructure.

You can now use Gemini 3.5 Flash in the Vercel AI SDK with built-in support for thinking levels and thought preservation. The AI Gateway provides automatic retries and reporting at no additional cost, following a pattern seen in GitHub Copilot's Gemini 3.5 Flash integration. This setup is suited for agentic tasks using Vercel's automated provider selection.

View the full update on vercel.com

Vercel Developers

@vercel_dev1d ago

Gemini 3.5 Flash is now on AI Gateway. Better coding, parallel agentic loops, and multi-turn reasoning. 𝚖𝚘𝚍𝚎𝚕: '𝚐𝚘𝚘𝚐𝚕𝚎/𝚐𝚎𝚖𝚒𝚗𝚒-𝟹.𝟻-𝚏𝚕𝚊𝚜𝚑' https://t.co/wLYAB98q81

760

View on X

Still wondering? A few quick answers below.

Gemini 3.5 Flash is Google's latest high-speed reasoning model now integrated into Vercel's AI Gateway. This integration allows developers to access the model's improved coding and multi-turn reasoning capabilities through a unified API. It eliminates the need for separate provider accounts while offering built-in observability and usage tracking.

Gemini 3.5 Flash replaces traditional randomness parameters with a thinking level configuration. It defaults to a medium setting to balance response quality with generation speed. Developers can adjust this level to optimize for complex reasoning tasks, which produces higher-quality reasoning traces that show the model's internal deliberation process.

When using Gemini 3.5 Flash through the Vercel AI SDK or AI Gateway, traditional sampling controls are not supported. This includes temperature, topP, and topK, as well as the thinking budget parameter. Instead, the model relies on its internal thinking configuration to manage the trade-off between reasoning depth and cost.

Vercel provides access to Gemini 3.5 Flash on the AI Gateway with no price markup over the base provider costs. Developers pay the standard rates while gaining infrastructure benefits like intelligent provider routing, automatic retries, and custom reporting. This makes it a cost-efficient option for high-volume agentic execution loops.

To use the model, you must set the model identifier to google/gemini-3.5-flash within your AI SDK configuration. You can then define provider options to set the thinking level and choose whether to include the model's internal thoughts in the output. This setup supports streaming text for real-time application responses.