Skip to content

Gemini 4 Hydrogen lands on Vercel's AI Gateway

Vercel added Google's new model to its unified model API under the name google/gemini-4-hydrogen, with claims of stronger agentic coding and fewer invented test results.

Draftship Editorial3 min read

Gemini 4 Hydrogen shows up on AI Gateway

On 12 November 2026, Vercel added Google's Gemini 4 Hydrogen to AI Gateway, its unified API for calling models from different providers. The model is reachable under the name google/gemini-4-hydrogen. For anyone already using the AI SDK, the switch is a one line change to the model string inside a streamText call, which Vercel shows in its changelog entry for the release.

Vercel describes Hydrogen as the successor to Gemini 3.8 Flash and says it stays in the same latency class while improving agentic coding, knowledge work and tool use. The post is short and carries no benchmark table, so everything about capability below comes from Vercel's own description rather than measured numbers.

What Vercel says the model does better

The claims cluster around long running agent work. Vercel says Hydrogen handles longer engineering tasks and produces better researched documents, spreadsheets and presentations in fewer agent steps. It also says the model follows system instructions, formatting requirements and task boundaries more closely, and that tool orchestration is more reliable across long conversations.

Two other areas are mentioned. Visual reasoning and technical problem solving have improved, which Vercel ties to multimodal workflows. And grounding is stronger, which it says reduces unsupported claims in search backed answers as well as invented tool calls or test results inside coding reports. That last point matters for anyone shipping agents that report back on their own work, because a fabricated test result is hard to catch from the outside.

How you reach it

There are two paths in the announcement. In code, point the AI SDK at the model name. In an existing coding agent, install the latest Vercel CLI and run npx vercel ai-gateway setup, then pick google/gemini-4-hydrogen inside the agent. Vercel names Claude Code, Codex and Cursor as the tools this covers, and points to its coding agents guide for the details.

AI Gateway itself is the surrounding layer: one API across models, usage and cost tracking, retries, failover and routing rules, per key budgets, Zero Data Retention support and custom reporting. Vercel says it mirrors provider pricing with no markup and charges no platform fee on inference, including bring your own key requests.

Limits and open questions

The announcement leaves out a lot of what a builder would normally check before switching. There is no context window, no token limit, no price per token for Hydrogen itself, and no rate limit information. Vercel gives no benchmarks, no eval results and no definition of what "same latency class" means in practice. It also does not say whether the model is generally available or in preview, or which regions can reach it. The grounding improvements are described qualitatively, with nothing independent behind them.

One practical caveat: the setup command is documented for Claude Code, Codex and Cursor by name. Whether it works the same way in other agents is not stated.

What it means for building apps with AI

Model choice is now a routine decision rather than a one time setup. If a gateway lets you swap the model string and keep the rest of your code, you can test a new release like Hydrogen against your own prompts without rewriting your app. That is the useful part here: the barrier to trying a model is a line of code or a CLI command.

For people building with an AI app builder, model options usually appear as a setting rather than a code change. Draftship, the AI app builder, generates and hosts an app from a plain language description, with a built in database, user sign-in and hosting, and its paid plans add a choice of AI model. If you are working that way, the guide to the AI chat and choosing a model covers how that selection works. Draftship has no affiliation with Google or Vercel, and nothing in this announcement says it supports Gemini 4 Hydrogen.

Primary sources

This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.