SecurityAlso reported elsewhere2 publishers2 min readPublished
Gemini 4 Argon launches at half the price Google plans to charge later
Google made Gemini 4 Argon widely available and says it ranks #1 on the Vals Index at 68.90%. On the CWE-bench v1 security benchmark it ties two other models at 68%.
The Watch · Security desk
Key figures: Argon input tokens cost $2 per million at the introductory price and $4 per million after the introductory period; output tokens cost $10 per million at the introductory price and $20 per million after.
| Figure | Value | Note | Claim |
|---|---|---|---|
| Input tokens, introductory price | $2 per million | 10 | |
| Input, after intro period | $4 per million | 11 | |
| Output, intro price | $10 per million | 10 | |
| Output, after intro period | $20 per million | 11 |
What happened
- Argon is open to developers and enterprises through the Gemini API, and to Google AI subscribers in Antigravity, the Gemini app and AI Mode in Search.
- On CWE-bench v1, which tests vulnerability remediation, Claude Opus 5.5 scores 67%, Claude Fable 5.1 scores 58% and Muse Spark 1.3 scores 55%.
- Google will release Argon without cyber guardrails to trusted defenders and its own internal teams.
- Google says the prices after the introductory period will be $4 and $20 respectively.
Why it matters
- capability The access criteria for trusted defenders decide who gets a model that Google says can find, validate and patch critical vulnerabilities on its own.
- decision Anyone running a bake-off between Argon and its rivals should score it at the post-introductory rate, since the launch rate is temporary.
- constraint The article does not include OpenAI or Anthropic prices or the date the introductory period ends, so the cost edge Google is pitching cannot be checked from it.
The Vals lead is narrow. The index weights agentic performance across domains by each sector's share of U.S. GDP [5]. Argon's 68.90% [2] sits 1.93 points above Claude Opus 5.5 at 66.97% [3][13] and 5.77 points above GPT-6 Astra at 63.13% [4][14]. Koray Kavukcuoglu, Google DeepMind's SVP and Chief AI Architect, said Argon is "ranking #1 on the Vals Index, which measures model performance across critical economic sectors like finance, coding, legal, and tax work" [12].
The cybersecurity case is thinner. The Hindustan Times article says Argon "seems to be ahead of its closest competition" on CWE-bench v1 [18]. Its own figures put Claude Opus 5.5 one point back [7][17], with two models level with Argon at the top [6].
On defense, Kavukcuoglu said: "Argon can autonomously find, validate, and patch critical software vulnerabilities." [8] Finding and validating a flaw is the same task for a defender and an attacker. The article does not say how a defender qualifies as trusted for the unrestricted version.
Google's stated bet is a mix of intelligence and lower costs [21]. At the introductory rates, cached input costs $0.10 per million tokens [15]. The later rates are exactly double the introductory ones on both input and output [16], so a workload sized on the launch bill costs twice as much at the same volume once the period ends.
The workflow evidence is Google's own. It cites a rise of more than 35% in agentic code submissions and a drop of nearly a third in code rollbacks for Argon in Antigravity [19]. It also cites a Chrome team job that split core webpage handling out of a roughly 6 million line legacy folder inside Chromium's 40 million-line codebase, completing months of work in weeks [20].
What to watch
- The end date of the introductory pricing period, since the $4 and $20 rates apply from that day.
- Published OpenAI and Anthropic prices for GPT-6 Astra and the Claude models, the only way to test Google's lower-cost claim.
- Independent runs of CWE-bench v1 and the Vals Index on Argon, and any criteria Google publishes for the version without cyber guardrails.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence40
- Adoption
- Insufficient
- Hype gap+30
- Incentives70
- Confidence55
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Argon is available globally to developers and enterprises via the Gemini API, and to Google AI subscribers in Google Antigravity, the Gemini app and AI Mode in Search.
- [2]
On the Vals benchmark, Gemini 4 Argon returns an accuracy of 68.90%.
- [3]
On the Vals benchmark, Claude Opus 5.5 scores 66.97%.
- [4]
On the Vals benchmark, GPT-6 Astra scores 63.13%, Meta Muse Spark 1.3 Max 58.16% and SpaceXAI's Grok 4.7 54.95%.
- [5]
The Vals benchmark tries to measure AI's potential economic impact by weighting agentic model performance across domains by each sector's share of U.S. GDP.
- [6]
On the CWE-bench v1 leaderboard, which evaluates a model's ability to remediate security vulnerabilities, Gemini 4 Argon ties at the top with 68%, alongside Grok 4.7 and GPT-6 Astra.
- [7]
On CWE-bench v1, Claude Opus 5.5 scores 67%, Claude Fable 5.1 scores 58% and Muse Spark 1.3 scores 55%.
- [8]
Koray Kavukcuoglu, SVP at Google DeepMind and Chief AI Architect at Google, said: "Argon can autonomously find, validate, and patch critical software vulnerabilities."
- [9]
For trusted defenders and Google's own internal teams, Google will release Argon without cyber guardrails.
- [10]
Argon's public access will be available at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off the input token price.
- [11]
After the introductory pricing period, Google says Argon's prices will be $4 and $20 respectively.
- [12]
Koray Kavukcuoglu said Argon is "ranking #1 on the Vals Index, which measures model performance across critical economic sectors like finance, coding, legal, and tax work".
- [13]
Argon's Vals score is 1.93 points above Claude Opus 5.5.
- [14]
Argon's Vals score is 5.77 points above GPT-6 Astra.
- [15]
At the introductory rate, cached input tokens cost $0.10 per million.
- [16]
The post-introductory prices are double the introductory prices on both input and output tokens.
- [17]
Argon's CWE-bench v1 score is one point above Claude Opus 5.5.
- [18]
The Hindustan Times article says Gemini 4 Argon 'seems to be ahead of its closest competition' in the CWE-bench v1 benchmark.
- [19]
Google cites that Argon in Antigravity is helping increase agentic code submissions by more than 35% while reducing code rollbacks by nearly a third.
- [20]
Google cites an internal example in which the Chrome team used Argon to separate core webpage handling from a roughly 6 million line legacy folder of browser features within Chromium's 40 million-line codebase, completing months of work in weeks.
- [21]
Google is betting on a combination of intelligence and lower costs to give it an edge in the economics of frontier AI.
Sources
2 independent publishers whose own reporting we read for this story.
- blockchain.newsGoogle Brings Gemini 4 Argon to Consumer App for AI Pro and Ultra Subscribers - Blockchain.News
2 articles · October 13, 2026
- hindustantimes.comGoogle takes aim in frontier AI race as Gemini 4 Argon leads in benchmarks | Business News
1 article · October 13, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.