Gemini 4 Argon
- Provider
- Status
- Restricted
- Context
- 1,000,000 tok
- Price
- $2 / $10 /MTok
Gemini 4 Argon is Google’s frontier model, announced on 30 September 2026 (Google). It is rolling out first to a set of trusted cyber defenders through Google’s Fairwind Program. Google says developers, enterprises and consumers follow “as soon as possible”, starting with paid API customers and Google AI Ultra subscribers, and has given no date.
Artificial Analysis scored it 53 on the Intelligence Index v4.3.2 at high reasoning, level with GPT-6 Astra and five points below Claude Opus 5.5 (58). Google announced an introductory price of $2/$10 per million tokens, rising to $4/$20, and raised the output limit to 1 million tokens.
Quick specs
| Gemini 4 Argon | |
|---|---|
| Announced | 30 September 2026 |
| Availability | Fairwind Program cyber defenders; public access not yet dated |
| Input price | $2.00 / MTok introductory, then $4.00 |
| Cached input | 95% off ($0.10 / MTok introductory) |
| Output price | $10.00 / MTok introductory, then $20.00 |
| Context window | 1,000,000 tokens |
| Max output | 1,000,000 tokens (up from 64,000) |
| Input | Text, images, video, audio |
| Knowledge cutoff | Not published |
| AA Intelligence Index | 53 (high) |
| Cost per AA index task | $1.99 at the introductory price |
| API model ID | Not published |
Who can use it
Now: Fairwind Program participants. Google says Argon is “rolling out to a set of trusted cyber defenders through our Fairwind Program”, the vetted-access scheme it opened on 2 September 2026 alongside Gemini 3.8 Flash Cyber. It says it will gather feedback from these early testers as it works on guardrails before a wider release.
Next: paid API customers and Google AI Ultra subscribers. Google names these two groups as the first public tier and gives no date beyond “as soon as possible”. Argon is not yet listed in Google’s Gemini API model or pricing documentation.
Government testing. Google says it is taking part in the US government’s voluntary process for pre-release model access.
Pricing
Google announced Argon at $2 per million input tokens and $10 per million output tokens as an introductory price, with cached input at 95% off. After the introductory period the price is $4 and $20. Google has not said when the introductory period ends.
The introductory price matches GPT-6.1 Sol and Claude Sonnet 5.5 ($2/$10). The standard price matches Claude Opus 5.5 ($4/$20).
Benchmark performance
Independent
| Measure | Gemini 4 Argon | GPT-6 Astra | Claude Opus 5.5 |
|---|---|---|---|
| Intelligence Index v4.3 | 53 (high) | 53 (max) | 58 (max) |
| Cost per index task | $1.99 | $3.26 | $5.98 |
| Terminal-Bench 4.0 | 57.1% | 59.1% | 59.6% |
| Hallucination rate (AA-Omniscience) | 15% | 51% | n/a |
| Price (in/out per MTok) | $2 / $10 intro | $10 / $50 | $4 / $20 |
Source: Artificial Analysis, with the component scores as reported by Trending Topics and The Decoder.
Intelligence. 53 on the index at high reasoning, which Artificial Analysis lists as #8 of 223 models: level with GPT-6 Astra at max effort, one point above GPT-6.1 Sol (52), three below Claude Sonnet 5.5 (56) and five below Claude Opus 5.5 (58).
Cost per task. $1.99 per completed index task at the introductory price, 60% of Astra’s $3.26. Artificial Analysis attributes the saving to token price, not token use: Argon averaged about 62,000 output tokens per task against Astra’s 27,000, and 110 million output tokens across the index against a median of 82 million. At the standard $4/$20, the same token use costs about twice as much.
Factuality. A 15% hallucination rate on AA-Omniscience, which Artificial Analysis says is the lowest of any model scoring above 45 on the index. Astra’s is 51%.
Agentic work. Artificial Analysis places it first on AutomationBench-AA at 77.5%, six points ahead of Claude Sonnet 5.5 (71.3%). On Terminal-Bench 4.0 it scores 57.1%, below Opus 5.5 and Astra.
Vals. Vals puts it first of 41 models on the Vals Index at 68.9%, ahead of Claude Sonnet 5.5 (67.0%), Claude Opus 5.5 (67.0%) and Claude Fable 5.1 (65.8%), and first on its Finance Agent v2 benchmark at 65.4%.
Arena. The Decoder reports it first on the Arena text leaderboard at 1,525 points.
Google’s own figures
| Benchmark | Gemini 4 Argon | Notes |
|---|---|---|
| DeepSWE v1.1 | 77.9% | Claude Opus 5.5 74.2%, GPT-6 Astra 74.1% on Google’s table |
| CWE-bench v1 | 68% | Vulnerability remediation; tied for first |
| AutomationBench | 51.3% | First |
| LVBench | 91.7% | Long video understanding |
Source: Google, vendor-run, with the competitor figures as reported by 9to5Google. Google also says Argon leads the Vals Index, Vals’s Finance Agent v2 and Harvey’s Legal Agent Benchmark. It published no SWE-bench Pro or SWE-bench Verified figure.
How it compares
vs GPT-6 Astra
GPT-6 Astra scores the same 53 on the index at $10/$50 and $3.26 per task, and is generally available. Argon costs $1.99 per task at its introductory price and has a hallucination rate of 15% against Astra’s 51%. Astra is ahead on Terminal-Bench 4.0 (59.1% to 57.1%) and uses fewer than half the output tokens per task.
vs Claude Opus 5.5
Claude Opus 5.5 scores 58 on the index, five points higher, at $4/$20 and $5.98 per task, and is generally available. Argon leads it on the Vals Index (68.9% to 67.0%) and on Google’s DeepSWE v1.1 table (77.9% to 74.2%).
vs GPT-6.1 Sol
GPT-6.1 Sol has the same $2/$10 token price as Argon’s introductory rate and scores one point lower (52), at $0.72 per completed task against Argon’s $1.99. It is on OpenAI’s API today.
vs Gemini 3.5 Pro
Gemini 3.5 Pro was announced at Google I/O on 19 May 2026 and never reached general release. Argon is Google’s first new frontier model since Gemini 3.1 Pro in February 2026, and The Decoder reports that 3.5 Pro was skipped.
Known limitations
Not publicly available. Access is limited to Fairwind Program participants. There is no public API, no Gemini app access and no date for either.
No SWE-bench figure. Google published neither SWE-bench Pro nor SWE-bench Verified, and no standardised harness has produced one.
Introductory pricing. The $2/$10 price doubles to $4/$20 after an introductory period whose end date Google has not given.
Verbose. Artificial Analysis measured about 62,000 output tokens per index task, more than twice GPT-6 Astra’s 27,000.
Documentation. No model card, knowledge cutoff or API model ID had been published at launch.
Frequently asked questions
When was Gemini 4 Argon released?
Google announced it on 30 September 2026 and began rolling it out the same day to trusted cyber defenders in its Fairwind Program. It has not been released to the public.
Can I use Gemini 4 Argon?
Not unless you are in Google’s Fairwind Program. Google says paid API customers and Google AI Ultra subscribers are next, “as soon as possible”, with no date.
How much does Gemini 4 Argon cost?
$2 per million input tokens and $10 per million output tokens at the introductory price, with cached input at 95% off. After the introductory period it costs $4 and $20. Google has not said when that period ends.
Is Gemini 4 Argon better than GPT-6 Astra?
Artificial Analysis scores the two level at 53 on its Intelligence Index. Argon costs $1.99 per completed task at its introductory price against Astra’s $3.26 and has a lower hallucination rate (15% to 51%). Astra scores higher on Terminal-Bench 4.0 and is generally available.
Is Gemini 4 Argon better than Claude Opus 5.5?
Not on Artificial Analysis’s Intelligence Index, where Opus 5.5 scores 58 to Argon’s 53. Argon is ahead on the Vals Index (68.9% to 67.0%) and on Google’s own DeepSWE v1.1 table.
What happened to Gemini 3.5 Pro?
It was announced in May 2026 and never reached general release. Google announced Gemini 4 Argon without shipping it.