Google's Gemini 4 Argon arrives on Vercel AI Gateway with 1M-token context

Google's Gemini 4 Argon model is now available on Vercel's AI Gateway, according to a Vercel blog post. The model is positioned for coding, agentic work and complex reasoning, with a 1 million token context window and input support for text, images, PDFs and video.

Built for multi-step coding work

According to Vercel, Gemini 4 Argon handles the tasks coding agents deal with every day: building features, fixing bugs and getting up to speed in unfamiliar codebases. The model is described as carrying longer tasks through multiple steps, which suits agentic workflows where a single job might involve reading code, calling tools and iterating toward a fix. Tool calling is supported natively.

The 1M token context window is the headline spec. For agent-style use, longer context means less chunking, less summarising and less retrieval scaffolding when a task touches many files or a large document set.

Vercel also points to the model's document handling: it reads dense diagrams, spreadsheets, layout-heavy documents and video, which the company suggests fits tasks like a financial analysis spread across several spreadsheet tabs. Beyond code, Vercel says it covers work in finance, health and law, as well as hard problems in math and science.

Thinking levels instead of budgets

A notable design choice: instead of thinking budgets, Gemini 4 Argon exposes thinking levels, low, medium and high. Rather than allocating a token budget for reasoning, developers pick a level. In the AI SDK this is set through the reasoning option; in the Chat Completions API it is reasoning_effort, with equivalent controls on the other supported APIs.

How to call it

The model name is google/gemini-4-argon. It can be reached through the AI SDK, the Chat Completions, Responses and Anthropic Messages APIs, or any coding agent connected to AI Gateway. For tools such as Claude Code, Codex and Cursor, Vercel says installing the latest Vercel CLI and running vercel ai-gateway setup is the path in; you then select google/gemini-4-argon inside the agent. The model is also available in Vercel's model playground for quick testing.

What AI Gateway adds

AI Gateway is Vercel's unified layer for calling language models. It provides a single API along with usage and cost tracking, retries, failover and performance optimisations that Vercel says deliver higher uptime than the underlying providers. Other features include built-in custom reporting, Zero Data Retention support, per-API-key budgets and routing rules.

Pricing follows the provider's own rates: Vercel charges no markup and no platform fee on inference, including on bring-your-own-key requests.

Why it matters

Two things stand out. First, the shape of the model: a 1M-token, multimodal model tuned for agentic coding is aimed squarely at how developers actually use AI today, in long tool-driven sessions rather than single prompt-response turns. The thinking-level control, which replaces fiddly token budgets, points in the same direction: simpler knobs for reasoning effort.

Second, the delivery route. A Google model that is callable through the Anthropic Messages API and usable inside Claude Code, Codex and Cursor after a one-time CLI setup shows what gateways are becoming: an abstraction layer over provider differences. For teams already routing traffic through AI Gateway, trying Gemini 4 Argon is a model-name change rather than an integration project, and with no markup on inference, the cost of experimenting is limited to the provider's own pricing.