Gemini 4 Argon
Google's new frontier model, announced at $2 / $10 introductory pricing with a 1M-token output limit; only trusted testers can use it so far.
What it is
Gemini 4 Argon was announced September 30, 2026 as Google's first new flagship since Gemini 3.1 Pro. Google is rolling it out to a set of trusted cyber defenders through its Fairwind Program first, then to paid API customers and Google AI Ultra subscribers; no date has been given. The headline spec change is the output limit: 1M tokens, up from 64K.
Pricing
| Input | $2.00 |
| Output | $10.00 |
| Cached input (cache read) | $0.100 |
| Cache write | not published |
| Batch input | not published |
| Batch output | not published |
| Fast / priority input | not published |
| Fast / priority output | not published |
| Standard price after introductory period | $4.00 / $20.00 |
- Introductory price $2 / $10; Google says $4 / $20 will apply after the introductory period. No end date published.
- Cached input: 95% off the input price.
- Batch, Flex, priority and grounding prices not published.
- Not listed on ai.google.dev pricing as of October 8, 2026.
Capability highlights
- Output limit raised from 64K to 1M tokens (vendor claim).
- Google reports state of the art on DeepSWE v1.1 (77.9%) and LVBench (91.7%).
- Positioned for software engineering, enterprise knowledge work (legal, finance) and cybersecurity defense.
- Access is limited to the Fairwind Program; developers cannot call it yet.
Modalities: text, image, video (LVBench reported) → text. Not published. Google's announcement states the output limit; it does not state the input context window.
How to use it
API: No public API model ID yet. Google says broader release will start with paid API customers and Google AI Ultra subscribers.
Versus Gemini 3.1 Pro Preview
Gemini 3.1 Pro lists at $2 / $12 (up to 200K tokens) on ai.google.dev; Argon's intro price is $2 / $10, rising to $4 / $20.
| Gemini 3.1 Pro Preview | Argon | |
|---|---|---|
| Input / 1M | $2.00 | $2.00 |
| Output / 1M | $12.00 | $10.00 |
| Cached input / 1M | $0.200 | $0.100 |
| Max output | 64K | 1M |
Published benchmark numbers
Vendor-reported, quoted from the linked announcement. Not independently verified; different vendors run benchmarks under different settings.
| Benchmark | Argon | Comparison in the same source | Source |
|---|---|---|---|
| DeepSWE v1.1 | 77.9% | GPT-6 Astra: 74.1%, Opus 5.5: 74.2% | Google DeepMind Gemini model page (Argon benchmark table) |
| Terminal-bench 4.0 | 57.4% | GPT-6 Astra: 58.2%, Opus 5.5: 66.4% | Google DeepMind Gemini model page (Argon benchmark table) |
| AutomationBench | 51.3% | GPT-6 Astra: 41.4%, Opus 5.5: 42.5% | Google DeepMind Gemini model page (Argon benchmark table) |
| GraphWalks 256k-1M BFS (F1) | 84.2% | GPT-6 Astra: 71.8%, Opus 5.5: 66.8% | Google DeepMind Gemini model page (Argon benchmark table) |
Comparisons
Sources
Every number above comes from one of these pages, read on the date shown. Searched for: gemini 4 argon, argon, gemini 4.
- Google blog: Gemini 4 Argon announcement read 2026-10-08
date, pricing, output limit, availability - Google DeepMind Gemini model page (Argon benchmark table) read 2026-10-08
benchmarks (vendor-reported) - Gemini API pricing (ai.google.dev) read 2026-10-08
Argon absent; Gemini 3.1 Pro price for comparison