<MCP.Playground/>
online

Gemini 3.7 Flash

Runs MCP tools

Google’s fast multimodal model for coding and agents, released 13 August 2026, three weeks after 3.6 Flash. Google pitches it against Claude Sonnet 5 and GPT-5.6 Terra at a lower price.

Vendor
Google
Released
13 August 2026
Context
1.05M tokens
Input
Text, image, video, audio, file

Paste a server URL, pick a model, and watch it call your tools in a real conversation. No install.

No server of your own? Leave it blank and use one of the public mock servers.

MCP Evals

Can Gemini 3.7 Flash actually complete tasks with your server?

One chat shows the tool calls work. An eval writes real tasks from your tool schemas, has Gemini 3.7 Flash drive each one, and reports where it got the wrong answer even though every call succeeded. Pick Gemini 3.7 Flash as the driver.

Run evals →See a sample

What Gemini 3.7 Flash is

Gemini 3.7 Flash is the Flash release between 3.6 Flash and 3.8 Flash, built for fast agent workflows, coding and multi-step reasoning. It has a 1.05M-token window, up to 64K tokens of output, and takes text, image, video, audio and file input. It is available through the Gemini API, AI Studio and Antigravity. Artificial Analysis scores it 56 on its Intelligence Index, four points above 3.6 Flash.

What Google says

  • FrontierCode 1.1: 43.6%, up from 34.4% for 3.6 Flash.
  • DeepSWE v1.1: 65.3%, up from 48.6%.
  • AutomationBench: 30.4%, up from 17% for 3.6 Flash.
  • 97.0% recall on GDM-MRCR v2 at 128K tokens.

These are the vendor’s own claims, not measurements of ours. Run the model against your own server to find out whether they hold for your tools.

How it reaches MCP

  • The Gemini API and Google’s Gen AI SDKs, which can take an MCP client session and turn its tools into function declarations
  • Gemini CLI, Google’s open-source terminal agent and an MCP client
  • MCP Playground’s Agent Studio: paste your server URL above and run 3.7 Flash against it in the browser, with no Google key

What to watch for

  • The AutomationBench jump from 17% to 30.4% is the number that matters most for MCP. It measures multi-step business workflows, which is what an agent chaining tool calls does. If 3.6 Flash dropped steps on your server, retest.
  • OpenRouter lists tools, tool_choice and structured_outputs but not parallel_tool_calls. A client that fans out several calls in one turn may see them come back one at a time.
  • Audio and video input do not reach the model through tool results. MCP tool content comes back as text or images, so the extra modalities only matter for what the user sends.

Run the same four-step task on 3.7 Flash and 3.8 Flash against the complex-schema mock: create_user_profile, process_order, analyze_data, then configure_workflow. If 3.7 Flash makes the same calls, you can stay on it; if it drops a later step, that is the gap 3.8 closes. All the mock servers →

Available here

ModelModel ID
Gemini 3.7 Flashgoogle/gemini-3.7-flash

Go deeper

Frequently asked questions

What changed from Gemini 3.6 Flash?
Google reports large gains on coding and agent benchmarks: FrontierCode from 34.4% to 43.6%, DeepSWE from 48.6% to 65.3%, and AutomationBench from 17% to 30.4%. It launched at half the price of 3.6 Flash.
Gemini 3.7 Flash or 3.8 Flash?
3.8 Flash is the newer model. Both have a 1M-plus window and the same request parameters on OpenRouter, so the quickest answer is to run both on your own server here and compare the calls.
Does Gemini 3.7 Flash support MCP?
Yes, through function calling. Google’s SDKs can pass an MCP session’s tools to the model, Gemini CLI is an MCP client, and any other client that turns MCP tools into function calls works too.
What is the context window of Gemini 3.7 Flash?
1,048,576 tokens of input, with up to 65,536 tokens of output, as OpenRouter lists it.

Sources

Gemini 3.7 Flash for MCP: Specs & Tool Calling — Test Free | MCP Playground