Google first unveiled Argon Sept 30, 2026, but limited it to trusted defenders in Fairwind Program while it completed US government voluntary pre-release safety checks. CEO Sundar Pichai posted on X: "We're going to make it available as soon as we can and as safely as we can." Today is that day.
What you actually get today
- API: Paid tier of Gemini API in AI Studio + Gemini Enterprise platform. Intro pricing: $2 per million input tokens, $10 per million output tokens, cached input 95% off. Price doubles to $4/$20 after Nov 30, 2026.
- App: Gemini app model picker now shows 3.5 Flash-Lite, 3.8 Flash, 3.1 Pro, and 4 Argon for Pro ($19.99) and Ultra ($99.99+) subscribers. Also in AI Mode in Search.
- Coding: Antigravity - Google's agentic IDE - gets Argon for Ultra subscribers via desktop app and CLI. For enterprise, eligible Gemini Enterprise subscriptions. Google also added toggle from Chat to Tasks, and macOS app now embeds Antigravity interface directly.
Early customers testing: Deloitte, LexisNexis, Ambience Healthcare, Isomorphic Labs. Google says 100k+ Googlers use internal Antigravity monthly.
Why Argon matters: deep work, not chat
Google built Argon for multi-step autonomous work, not single-turn Q&A. Output limit jumped from 64K to 1 million tokens, so it can refactor a huge codebase in one run.
Benchmarks - vendor-run, so caveat:
- Vals Index (finance, coding, legal, tax weighted by GDP share): Argon #1 at 68.90% as of Oct 12, ahead of Claude Sonnet 5.5 67.04% and Opus 5.5 66.97%
- DeepSW v1 long-horizon coding: 77.9% vs Claude Opus 5.5 74.2% vs GPT-6 Astra 74.1%
- Legal Agent banchmark (Harvey): 19.6% vs Astra 5.4%
- CWE-bench v1 security patching: tied with Astra at 68%
It still trails Astra on FrontierSWE v2 (65.5% to 55%) and OSWorld-2.0, and Opus on Terminal-bench 4.0.
Internal win Google cites: Antigravity teams increased agentic code submissions 35%+ and cut rollbacks by nearly a third over 4 weeks. One demo: Argon agents freed 300 TiB memory across data centers by optimizing allocations, projected 500 TiB to 1 PiB total.
Google describes Argon style: answers more directly, less hedging, skips clichés, follows style instructions more reliably, and narrates progress on long tasks.
Context: come back after rough stretch
This launch follows cancellation of Gemini 3.5 Pro promised for June, and DeepMind overhaul after co-founder Demis Hassabis stepped aside as CEO to Chair in August and several Gemini leads left. Google had shifted messaging to cost advantage; Argon puts it back in frontier race vs OpenAI GPT-6 Astra and Anthropic Claude Opus 5.5.
Fairwind now opens to hundreds more partners, giving defenders a version without cyber guardrails to hunt vulns. Google also offering 10,000 scientists 6 months free Antigravity top tier.
FAQ
Is Gemini 4 Argon free?
No. Requires paid API or Google AI Pro/Ultra. Pro is ∼$19.99/mo, Ultra ∼$99.99+/mo. No free tier for Argon yet.
How to access Gemini 4 Argon in API?
In AI Studio, select `gemini-4-argon` in model picker under paid tier. Via Vertex AI Gemini platform same. Enterprise via Gemini Enterprise.
What is Google Antigravity?
Agent-first IDE + CLI from Google for long-running coding agents. Now has Argon inside for Ultra. Lets agent plan, edit, run tests, open PRs autonomously.
Is Gemini 4 Argon better than GPT-6 Astra?
On Vals Index and DeepSWE, Argon leads Google's tests. Astra leads on FrontierSWE v2 and OSWorld. Expect tradeoffs. Wait for independent evals.
What does Argon name mean?
Argon is first in Gemini 4 series, codename for noble gas - inert, stable - fitting for deep work + cybersecurity focus.
Also read this: https://afterpostint.com/post/p-strong-hackers-turn-french-supermarket-freezers-into-50-c-ovens-strong-p