Google's new frontier model for long-running, multi-step work — $2/$10 per 1M tokens through Nov 30, 2026, live in the Gemini app for AI Pro/Ultra and via the API for every developer.
Gemini 4 Argon bills $2 / 1M input and $10 / 1M output through Nov 30, 2026 as an introductory rate. Cached input is $0.10 / 1M. The rate reverts to a standard $4/$20 (cached $0.20) on Dec 1, 2026. It is not available on the free tier.
Google unveiled Gemini 4 "Argon" on Sep 30, 2026 as its answer to OpenAI's GPT-6 Astra — a new frontier model built for long-running, multi-step tasks in software engineering, enterprise knowledge work, finance, law, science and cyber defense. At launch it was limited to trusted cyber-defense partners through the Fairwind Program. Google then opened it up fast: it reached general availability on Oct 13, 2026, landing in the consumer Gemini app for Google AI Pro and AI Ultra subscribers (and AI Mode in Search), and via the API under the model id gemini-4-argon for every developer. AI Ultra subscribers additionally get it in AI Studio and the Antigravity coding app.
The intro pricing window is temporary: Google's own pricing page lists Dec 1, 2026 as the date rates double to a standard $4/$20 per 1M (cached input $0.20) — the same pattern Google used for recent Flash-tier launches. Budget workloads that run past the new year accordingly.
Argon supports "long-decode" generation up to a 1M-token context, which Google is positioning for tasks that unfold over many steps rather than a single response — multi-file coding changes, drawn-out agentic workflows, and research synthesis across large document sets. Only implicit prompt caching is supported; there's no explicit cache-management API like some rival models offer.
Access is still being rolled out gradually — Google says the model may show up in a given account's picker a few hours after the general-availability date, so a brief delay on launch day isn't a bug.
| Model | Input / 1M | Output / 1M | Context |
|---|---|---|---|
| Gemini 4 Argon (intro) | $2 | $10 | 1M |
| Gemini 3.1 Pro | $2 | $12 | 1M |
| GPT-6 Astra | $10 | $50 | 1M |
| Claude Fable 5.1 | $10 | $50 | 1M |
| Claude Opus 5.5 | $4 | $20 | 1M |
At its introductory rate, Argon undercuts every other current flagship on output price while matching 3.1 Pro's input — making it, for now, the cheapest way to try a true frontier-class model from any major lab. That changes Dec 1, 2026 when it doubles to $4/$20, landing right alongside Claude Opus 5.5.