Google Unveils Gemini 4 Argon, Its Most Powerful AI Model Yet
Translated & summarized from Behadrei Haredim by baba
The story in 6 lines · by baba
- Google unveiled Gemini 4 Argon, its most powerful AI model, designed for complex tasks.
- Argon boasts a one million token output limit, a major increase from previous models.
- The model shows strong performance in benchmarks, particularly in cybersecurity and legal domains.
- Israeli cybersecurity firm Wiz is already using Argon to find vulnerabilities in public infrastructure.
- Google is implementing a phased rollout, starting with trusted partners before public release.
- Initial pricing is set at $2 per million input tokens and $10 per million output tokens.
Google DeepMind has launched Gemini 4 Argon, the flagship model of its next-generation Gemini AI series. Designed for extensive and complex tasks in software development, finance, law, and cybersecurity, Argon is initially not available to the general public. A key advancement is its output limit of up to one million tokens, a significant increase from the 64,000 tokens of the previous generation. Google states this allows the model to process lengthy tasks, analyze and generate vast amounts of code, and execute complex operations sequentially.
In performance benchmarks released by Google, Argon achieved 77.9% on DeepSWE v1.1, surpassing GPT-6 Astra's 74.1% and Claude Opus 5.5's 74.2%. It also led the Vals Index with 68.9% and scored 19.6% on Harvey's legal test. However, it did not top all categories, with Claude Opus 5.5 scoring 66.4% on Terminal-Bench 4.0 compared to Argon's 57.4%, and GPT-6 Astra leading OSWorld-2.0 with 72.6% against Argon's 69.2%.
The Israeli connection comes through Wiz, a cybersecurity firm founded by Israeli entrepreneurs. Wiz is already utilizing Argon in its Scan for Good initiative, which identifies vulnerabilities in public infrastructure. Google reported that Argon successfully detected a critical weakness in medical software used by hospitals worldwide, a flaw missed by previous leading models. Google itself is also using Argon internally, for optimizing quantum calculations, saving hundreds of terabytes of memory in data centers, and migrating large code segments from C and C++ to Rust, including work on the Zircon core of the Fuchsia operating system.
Despite its capabilities, Google has opted for a limited rollout. Argon is first accessible to a select group of trusted cybersecurity entities through the Fairwind program, while the company continues safety testing and participates in a voluntary U.S. government review process. Google plans to subsequently offer it to paying API customers and Google AI Ultra subscribers before wider release. Initial pricing is set at $2 per million input tokens and $10 per million output tokens, with a 95% discount for cached input. Post-introductory period, prices are expected to rise to $4 for input and $20 for output.
Read the original at Behadrei HaredimThe same event, reported separately by each outlet. Open a few to compare what different newsrooms emphasize — and what they leave out.
