Gemini 3.7 Flash lands on Vercel AI Gateway with a 50% discount

Google's Gemini 3.7 Flash has landed on Vercel's AI Gateway, and for the next couple of years you can run it at half price. The discount runs until December 31st, 2026, which is a long window for a model launch — a sign Vercel is serious about making this the default cheap-and-fast option for agentic workloads.

The model is positioned as a step up from earlier Flash variants, specifically for software engineering and agentic tasks. Vercel's blurb highlights two concrete improvements: it resolves issues more reliably and spends less time stuck in failed agent loops. That second point matters more than it sounds — on long tool-calling sequences, one derailment can waste the entire run's budget. If Gemini 3.7 Flash genuinely cuts those failure spirals, it's a meaningful win for anyone running autonomous coding agents.

There's also a design-to-code angle: the model can generate desktop and web application code directly from design mocks, with closer adherence to the source design. That's a crowded space, but tighter fidelity to the mock is the kind of thing that separates a useful tool from a demo.

Getting started is straightforward. In the AI SDK, set the model to google/gemini-3.7-flash:

import { streamText } from 'ai';
const result = streamText({
  model: 'google/gemini-3.7-flash',
  prompt: 'Build a settings page from this design mock.',
});

For coding agents, Vercel's CLI wires it up too: run vercel ai-gateway coding-agents setup to connect Claude Code, Codex, OpenCode, or Pi, then select google/gemini-3.7-flash inside the agent. No-code folks can poke at it in the model playground.

AI Gateway itself remains a unified API for calling models, with usage and cost tracking, retries, failover, and performance optimizations. It also supports custom reporting, Zero Data Retention, API key budgets, and routing rules. Vercel continues to claim no markup on inference and no platform fee, including for BYOK requests — that's the part that keeps the gateway honest as a pure proxy rather than a toll booth.

If you're already on the AI Gateway, this is a low-friction way to test a new model without changing your integration. The 50% discount sweetens the deal, but the real question is whether the reliability improvements hold up in production. That's something benchmarks won't tell you.