Gemini 4 Argon: What Changed, What It Costs, Should You Switch?
Answer first
Gemini 4 Argon reached general availability on October 13, 2026: Google's frontier flagship with a 1M-token window, multimodal input, and introductory API pricing of $2/$10 per million tokens that doubles on December 1. Independent Artificial Analysis scoring puts it at 53 on the Intelligence Index — tied with GPT-6 Astra, behind Claude Opus 5.5's 58.
What changed
- A staged release: announced September 30, 2026 to vetted cyber defenders in Google's Fairwind Program; generally available October 13 on the paid Gemini API and for Google AI Pro/Ultra subscribers. It is not on any free tier.
- Google's first frontier flagship in over seven months (per MetricNexus's launch analysis) — this is a catch-up release, and it lands mid-pack, not on top.
- Output ceiling: headline max output is 1M tokens, up from 64K — with a call-path catch (see "Runs on what?").
- One shared window: input, thinking and output all draw from the same 1,048,576-token budget rather than separate allowances (per MetricNexus).
- Style change, per Google: answers more directly, with less hedging and repetition.
- Multimodal input: text, images, video and audio.
The numbers, with trust labels
| Benchmark | Gemini 4 Argon | Comparison | Source & trust |
|---|---|---|---|
| DeepSWE v1.1 | 77.9% | Claude Opus 5.5: 74.2% | Google's launch table — vendor-reported (via VentureBeat / Lindy) |
| FrontierSWE v2, Terminal-Bench 4.0, PostTrainBench, Terminal-Bench Science | trails Opus 5.5 | Opus 5.5 ahead on all four | Google's launch table — vendor-reported (via Lindy) |
| Intelligence Index v4.3.2 | 53 | GPT-6 Astra 53; Claude Sonnet 5.5 56; Claude Opus 5.5 58 | Artificial Analysis — independent (via Fello AI; MetricNexus) |
| Cost per Intelligence Index task | $1.99 at introductory price; $3.98 at standard price | GPT-6 Astra $3.26 | Artificial Analysis measurement — independent (via MetricNexus) |
How to read this table: Google's own table has Argon ahead of Opus 5.5 on most rows — and the independent index has Opus 5.5 ahead overall, 58 to 53. Both can be true: vendor tables pick favourable tasks, indexes average everything. The durable facts are the tie with GPT-6 Astra on the independent index and the cost-per-task swing: at the introductory price Argon is the cheapest way to buy this tier of intelligence; at the standard price it costs about 1.2× Astra per task (MetricNexus).
What it costs (per 1M tokens, checked Oct 23, 2026)
| Route | Input | Output | Cached input | Note |
|---|---|---|---|---|
| Introductory API price (through Nov 30, 2026) | $2.00 | $10.00 | $0.10 (95% off) | Per Google's launch announcement and API pricing page |
| Standard API price (from Dec 1, 2026) | $4.00 | $20.00 | — | Exactly double — budget on this number |
| Subscription access | — | — | — | Gemini app and AI Mode for Google AI Pro ($19.99/mo) and Ultra subscribers; not on the free app |
Runs on what?
- Closed and paid-only. No weights, no free tier. Routes in: the paid Gemini API (model ID
gemini-4-argon) and AI Studio; the Gemini app for Pro/Ultra subscribers; Antigravity, Google's agentic coding environment, requires Ultra for individual subscribers. - The 1M output headline has fine print: per MetricNexus's testing notes, a normal synchronous request still stops at 65,536 output tokens — reaching the headline limit takes a specific call path. Verify against your integration before promising anyone a 1M-token answer.
- No published parameter count or knowledge cutoff (see below) — Google is shipping this as a service, not a spec sheet.
Should you switch? By situation
- You already build on Google's stack (Gemini API, AI Studio, Antigravity) → Evaluate now, and use the window. The six weeks of introductory pricing are the cheap period to prototype and measure real token usage before the December 1 doubling.
- You are choosing a frontier API on cost per task → Argon, until Nov 30. At $1.99 per index task it undercuts Astra ($3.26); after the doubling it is the more expensive of the two per task.
- You want the strongest frontier model regardless of vendor → Not Argon today. Opus 5.5 leads the independent index (58 vs 53) and several rows of Google's own table, including the terminal/agent benchmarks.
- You are on a free Gemini tier → No access. Argon requires a paid API account or a Pro/Ultra subscription.
FAQ
When was Gemini 4 Argon released?
Google announced it on September 30, 2026, initially only for vetted cyber-defense partners in its Fairwind Program. General availability followed on October 13, 2026, for paid Gemini API customers and Google AI Pro and Ultra subscribers.
Is Gemini 4 Argon free to use?
No. There is no free tier for Argon — not in the Gemini app and not on the API. The cheapest routes in are pay-as-you-go API billing at the introductory $2/$10 rate, or a Google AI Pro subscription at $19.99 per month for app access.
Is Gemini 4 Argon better than GPT-6 Astra?
On Artificial Analysis's independent Intelligence Index they tie at 53. Google's own benchmark table favours Argon on most rows it publishes, while Astra wins several agent and terminal rows in that same table. Treat them as the same tier and decide on price, ecosystem and your own evals.
Why does the price double on December 1?
The $2/$10 rate is introductory launch pricing, announced by Google as running through November 30, 2026. The standard rate from December 1 is $4/$20 per million tokens. Any budget written after November should use the standard rate.
Can Gemini 4 Argon really output a million tokens?
The headline limit is real but path-dependent: MetricNexus found that a standard synchronous API request still caps output at 65,536 tokens, and the 1M figure is reached through a specific call path, with input, thinking and output sharing one window. Check the current API documentation for your route before relying on it.
What we could not verify
- Parameter count and knowledge cutoff: not published by Google in the sources checked
- LMArena standing: one research roundup reported an Arena Text Elo of 1,525 (leading in early October) — single-source, not confirmed against LMArena's own leaderboard in the sources checked
- Whether the 1M-output call path is available on all API routes and regions
Sources: Google launch announcement and API pricing (as covered by CoinStats and Lindy's launch guide, Oct 13, 2026, checked Oct 23, 2026); Artificial Analysis Intelligence Index v4.3.2 figures via Fello AI and MetricNexus (Oct 2026); Gemini API release notes via TechAIWire; The AI Rankings spec sheet. Vendor-reported figures are labelled as such wherever they appear on this card.