Google DeepMind officially expanded its flagship AI model Gemini 4 Argon on October 13, making it available to global developers, enterprises, and Google AI Pro and Ultra subscribers. When this model debuted on September 30, it was only available to approved security defense partners through the Fairwind Program. Now it has entered the general commercial application market. Google simultaneously launched a limited-time API discount, charging $2 for each million input tokens and $10 for output, which is half the official price starting in December. Gemini 4 Argon is mainly designed for work that requires long-term reasoning and multi-step processing, including large-scale code modification, financial and legal document analysis, information security defense and scientific research. The model supports a unified context window of up to 1,048,576 tokens, which is shared by input, inference, and output. Developers can handle extremely long tasks through a special continuation generation mechanism, but the default output limit of general synchronous API calls is 65,536 tokens. To generate longer content, additional settings must be made.
In terms of individual users, Google AI Pro and Ultra subscribers can already choose Argon in Gemini App and Google Search AI Mode. Ultra members can also use the model for program development through Google Antigravity. Free version and Google AI Plus users are not yet included in the direct open scope. Enterprises can use Argon from services such as Google Cloud and Gemini for business. Developers can connect through Google AI Studio or Gemini API. The model identification code is "gemini-4-argon".
Please read on...
Google has set this API promotion until November 30, during which it will charge US$2 and US$10 per million input and output tokens respectively. If the request meets the implicit cache conditions and is successfully hit, entering the token can enjoy a 95% discount, down to $0.1 per million. Starting from December 1, the standard price will be adjusted to US$4 for input and US$20 for output, which is the same as the public pricing of Anthropic Claude Opus 5.5; OpenAI GPT-6 Astra is US$10 for input and US$50 for output.
The lower price also puts the Argon's review performance in the spotlight. The Intelligence Index released by Artificial Analysis, a third-party organization, shows that Gemini 4 Argon scored 53 points under the High inference setting, which is the same as the Max setting of GPT-6 Astra and lower than the 58 points of Claude Opus 5.5. This comprehensive review doesn’t show Argon outperforming its competitors across the board, but its API unit token pricing was significantly lower during the launch promotion period.
Google's self-published tests showed mixed results. In the Vals Index, which covers professional jobs such as finance, law, program development and taxation, Argon achieved 68.9%, higher than Claude Opus 5.5's 67.0%, Claude Fable 5.1's 65.8% and GPT-6 Astra's 63.1%. Program development test DeepSWE v1.1 is also led by Argon with 77.9%, Opus 5.5 and Astra with 74.2% and 74.1% respectively. However, if we look at terminal operation and actual software engineering issues, the results are different. Terminal-Bench 4.0 was led by Opus 5.5 with 66.4% and Argon with 57.4%; in FrontierSWE v2, Astra achieved 65.5%, which was better than Opus 5.5's 62.3% and Argon's 55.0%. Argon has good results in tests such as financial agency tasks and AutomationBench, but it has not yet achieved comprehensive advantages in software engineering-related projects. The above data comes from comparative data published by Google and should be interpreted separately from the independent evaluation of Artificial Analysis.
Google also announced the results of its internal development work using the model. According to official statistics, after the engineering team implemented AI agent program development through Argon, the number of code submissions increased by more than 35%, and the proportion of code submissions that needed to be rolled back was reduced by nearly 30%. The Chrome team has also used Argon to handle re-architecting work across hundreds of files. However, these performance data mainly come from Google's own deployment experience and may not be directly applicable to other enterprise environments.
In addition to enterprises and paying users, Google is also preparing to provide 10,000 qualified scientists with six months of free quota at the highest usage level of Google Antigravity, which will be opened in batches. The Fairwind Program, which was originally limited to security defense partners, has also expanded its recruitment to allow more vetted research and defense institutions to use special versions of Argon.
