Build2 publishersIndependently confirmed2 min readPublished
Gemini 4 Argon's API price doubles when Google's introductory window ends
Google announced Gemini 4 Argon, its first Gemini 4 model, with early access limited to internal teams and vetted cyber defenders in its Fairwind program. The published API price doubles once an introductory window closes. Teams cannot yet test the coding and long-output claims they would budget around, because there is no public access.
The Engineer · Build desk
What happened
- Argon is the first model in the Gemini 4 line and the first new frontier Gemini that Google has shipped since 3.1 Pro.
- Google skipped a 3.5 Pro release over the summer and shipped 3.8 Flash instead.
- Google aims Argon at real-world software engineering, legal and finance knowledge work, and defensive cybersecurity that finds and patches vulnerabilities.
- Google says Argon leads the Vals Index, a benchmark that scores professional work across several fields.
- Architecture, parameter count, active parameter count, and knowledge cutoff are all unpublished.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- constraint The only performance numbers available are Google's own, so a team cannot test Argon's coding or long-output claims before it commits budget to them.
- cost A team whose production rollout lands after the introductory window pays the higher rate on every token, so budgets built on $2/$10 understate the real bill.
- contradiction Runtimewire calls the post-intro price '2.5 times higher,' but each published rate doubles exactly, so the 2.5x figure is not supported by the stated numbers.
- capability If the 1M-token output holds up outside Google, a migration or long analysis could run in one response instead of being split across many calls.
The introductory rate is $2 per million input tokens and $10 per million output [9]. When the window closes, the stated price is $4 and $20 [10]. Both figures are exactly twice the introductory ones [18]. Runtimewire describes the increase as "2.5 times higher" [2], which does not follow from these rates: double the input price and double the output price, and any mix of the two still comes out at twice the introductory bill. The window has no date and the API is not open, so no team outside the program can measure its own token mix against either number yet [1][6].
The numbers a buyer would plan around are all Google's own. They come from a September 30th post and a chart Sundar Pichai posted, several rivals were scored by Google, and independent replication is not out because almost no one outside Fairwind can run the model [3]. Google says Argon beats its 3.8 Flash Cyber model on an internal vulnerability-discovery set across 20 languages and on Wiz's black-box web pentest benchmark, and Wiz is an early Fairwind deployer [15]. Nate's Newsletter put the access problem plainly: "most of us still can't use it ourselves" [5].
Google's examples of long-horizon work are internal and unaudited [22]. It says an Argon Rust port of libgav1 replaced 32,000 lines of SIMD and runs 2.7x faster than the existing Rust port with identical video output, without stating the comparison against the original C++ [19]. It says agents covered more than 800,000 lines of the Fuchsia Zircon kernel in a C and C++ to Rust migration still under automated and manual audit [20], and freed over 300 TiB of memory from server telemetry [21]. Argon's output ceiling is up to 1 million tokens in a single response, against 64K on earlier Gemini models [11].
Several things a buyer would want closed are still open. Input context was not clearly specified at launch, and only some trackers list 1M [12]. Google says broader access is planned [6] but names the paid API and Google AI Ultra as the next tier without a date [16]. Rumors of access around October 9th or 10th are not confirmed by any Google account [17].
What to watch
- A model card or API changelog landing, which would let anyone replicate Google's benchmark figures.
- A dated opening of the paid API and Google AI Ultra tier, which Google has named but not scheduled.
- Whether input context is fixed at 1M or made selectable, which the launch post did not clarify.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence35
- Adoption8
- Hype gap+40
- Incentives65
- Confidence50
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
As of October 8th 2026, Argon is announced but not generally available; confirmed access is Google internal teams plus vetted cyber defenders in the Fairwind Program.
- [2]
Runtimewire describes the post-introductory pricing as 2.5 times higher than the introductory rates.
ReportedSupportedSource: runtimewire.com2 sources— create a free account to open themView cited source - [3]
All published benchmark figures are Google's own, from a September 30th post and a chart Sundar Pichai posted; several rivals were scored by Google, and independent replication is not out because almost no one outside Fairwind can run the model.
- [4]
Google says Gemini 4 Argon leads the Vals Index, which evaluates professional work across several fields.
ReportedSupportedSource: Google, per natesnewsletter.substack.com2 sources— create a free account to open themView cited source - [5]
most of us still can't use it ourselves
ReportedSupportedSource: natesnewsletter.substack.com2 sources— create a free account to open themView cited source - [6]
The initial rollout is limited to trusted cybersecurity defenders, with broader access planned.
- [7]
Argon is the first Gemini 4 model and the first new frontier Gemini since 3.1 Pro.
- [8]
Google skipped 3.5 Pro over the summer and shipped 3.8 Flash instead.
- [9]
Introductory API pricing is $2 per million input tokens and $10 per million output tokens.
- [10]
Stated post-introductory API pricing is $4 per million input tokens and $20 per million output tokens.
- [11]
Argon can generate up to 1 million tokens in one response, up from 64K on earlier Gemini models.
- [12]
Input context was not clearly specified in the launch post; some trackers list 1M input.
- [13]
Architecture, parameter count, active parameters, and knowledge cutoff are unpublished.
- [14]
Google names three jobs for Argon: real-world software engineering; enterprise knowledge work, especially legal and finance; and defensive cybersecurity.
- [15]
Google says Argon beats its 3.8 Flash Cyber model on an internal vulnerability-discovery set across 20 languages and on Wiz's black-box web pentest benchmark; Wiz is an early Fairwind deployer.
- [16]
Paid API and Google AI Ultra are the named next tier, which Google has not dated.
- [17]
Rumored paid API and Ultra access around October 9th or 10th is not confirmed by any Google account.
- [18]
Each published per-token rate multiplies by exactly 2.0 from the introductory to the post-introductory pricing, so the stated increase is a doubling rather than 2.5 times.
- [19]
Google says an internal libgav1 Rust port replaced 32,000 lines of SIMD and runs 2.7x faster than the existing Rust port with identical video output; the comparison to the original C++ is not stated.
- [20]
Google says Argon agents migrating C and C++ to Rust covered up to 800,000+ lines of the Fuchsia Zircon kernel, still under automated and manual audit before production.
- [21]
Google says Argon agents freed over 300 TiB of memory from server telemetry, with 500 TiB to 1 PiB projected once rolled out.
- [22]
Google's internal results are unaudited and not yet evidence of results outside the company.
Sources
2 independent publishers whose own reporting we read for this story.
- natesnewsletter.substack.comGoogle says Gemini 4 Argon tops the Vals Index. Your customers will grade you on something else: the job they could never justify doing themselves. Here’s how to find it.
1 article · October 7, 2026
- runtimewire.comEverything we know about Google's Gemini 4 Argon
1 article · October 8, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Topics
- AI for CybersecurityFollow
- AI BenchmarksFollow
- LLM API PricingFollow
- Frontier Model Access GatingFollow