Gemini 4 Argon Reached General Availability October 13 at a $2 Promotional Price, After a Cyber-Partners-First Launch
A.I. / news
Gemini 4 Argon Reached General Availability October 13 at a $2 Promotional Price, After a Cyber-Partners-First Launch
Google's own post gives four benchmark scores, concedes two coding tests to rivals in Reuters' account, and sets a $4 input price from November 30.
Google made Gemini 4 Argon generally available in the Gemini API on October 13, 13 days after it announced the model to a small group of cyber defenders. The Gemini API changelog lists the model ID as gemini-4-argon.
Google announced Argon on September 30 in a post by Koray Kavukcuoglu, senior vice president of Google DeepMind and Chief AI Architect. At that point it was rolling out through the Fairwind Program to trusted cyber defenders, with broader access promised for paid API customers and Google AI Ultra subscribers.
What Google claims
The post calls Argon Google's frontier model for real-world coding, enterprise knowledge work and cyber defense. All scores below are Google's own.
| Benchmark | Google's score | Google's description |
|---|---|---|
| DeepSWE v1.1 | 77.9% | State of the art |
| AutomationBench (Zapier) | 51.3% | Ranked first |
| LVBench | 91.7% | State of the art |
| CWE-bench v1 | 68% | Tied for first |
The post says Argon can find, validate and patch critical software vulnerabilities on its own, and that it is Google's most resilient model against indirect prompt injection. It gives no score for the Gray Swan prompt-injection benchmark or for the Vals Index, only the word "leading".
- LVBench91.7 %
- DeepSWE v1.177.9 %
- CWE-bench v168 %
- AutomationBench51.3 %
Source: Google blog post by Koray Kavukcuoglu, September 30, 2026 (vendor-supplied)
What Google did not lead on
Reuters, in a story published October 1, reported that Argon trailed rivals on two of the four coding benchmarks Google included in its release. Google said Argon is comparable to OpenAI's Astra and Anthropic's Opus on key coding and cyber benchmarks, and a spokesperson said it is "our most performant model yet built for complex workloads."
Reuters also reported that Google no longer plans to release Gemini 3.5 Pro, which Sundar Pichai had said in June would ship, and that Demis Hassabis stepped aside as DeepMind's founder and chief executive in an overhaul of the lab.
Price and limits
Google's introductory price is $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after the introductory period, with cached input 95% off the input price. A secondary report puts the change on November 30. The $4 and $20 rate matches the list price Anthropic set for Opus 5.5 on September 22. On the open-weight side, DeepSeek's V4.1-Flash-1017 is the release the price cuts are being measured against.
The post says the output token limit rises to 1 million tokens, up from 64,000. It does not state a context window figure, and the changelog entry that The Terminal read carries no pricing.
The cyber-first release follows a pattern seen in the Cloudflare security audit skill coverage, where vendors pitch agents at vulnerability finding. Google did not publish a date for the end of the introductory rate on its own post.
Sources disagree on which subscription tiers get Argon inside Antigravity, and Google's model table for that product is the authority.
Sources
More in A.I.
- 01Qwen3.8-27B Ships Under Apache 2.0 With 262,144-Token Context, While Its 2.4T Sibling Carries a $50 Million Revenue ClauseThe model card lists 25 benchmark rows, all Alibaba's own, and says nothing about training data or the hardware needed.
- 02OpenAI Cut Luna 80% in One Step. Anthropic Has Cut 20% at a Time.Since June 26 both labs have repriced their model lines, and the dated record shows two different methods, with DeepSeek still far below either.
- 03Microsoft's Decision-1 Is a Post-Trained Qwen3.5-9B at $0.042 Per Million Tokens, With Self-Reported BenchmarksThe model scores fixed answer options instead of writing text, and the launch post does not say whether the weights can be downloaded.
- 04Eka's 1,500 Picks an Hour Is a Lab Number, and Its Own Investors Say SoThe Cambridge, Mass. startup raised $150 million for a vision-force-action robot model. Its headline throughput comes from induction and bin-picking tests, and customer pilots are unnamed.