Google has unveiled Gemini 4 Argon, its new flagship AI model, but it is not yet broadly available. Rollout has begun with a set of trusted cyber defenders through the Fairwind Program. The company says wider access for developers, enterprises and consumers will follow 'as soon as possible', beginning with paid API customers and Google AI Ultra subscribers. No general availability date has been given.
Gemini 4 Argon: Key features
Google describes Argon as built for long, multi-step work in software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defence. The headline spec is a sharply higher output ceiling: 1 million output tokens, up from 64,000 previously, which lets the model reason and generate across very long tasks in a single run.
On Google's published benchmarks, Argon scores 77.9 percent on DeepSWE v1.1 (real-world software engineering), 51.3 percent on Zapier's AutomationBench and 91.7 percent on LVBench for long-video understanding. It ties for first on CWE-bench v1, a vulnerability-remediation test, with 68 percent. Google also says the model can autonomously find, validate and patch critical software flaws.
Internally, thousands of Google employees already use Argon. In one example, agents analysing fleet-wide telemetry freed up over 300 TiB of memory across data centres. Google also cites a rewritten libgav1 video decoder that runs 2.7x faster than its earlier Rust port.
Gemini 4 Argon: Why it matters
The launch comes at a sensitive time for Google. The company had not released a new flagship since the Gemini 3 series in November 2025, while OpenAI and Anthropic moved ahead with GPT-6 and newer Claude models. Alphabet also faced investor concern over a delayed Gemini 3.5 Pro and a leadership reshuffle at DeepMind. Argon is Google's strongest bid yet to regain frontier standing, with a focus on the areas where enterprises are already spending.
Cybersecurity is the other major theme. Google says trusted defenders and its own teams will get Argon without cyber guardrails, so they can use its full defensive capability. Security firm Wiz, using Argon through its Scan for Good initiative, found a critical vulnerability exposing personal data in healthcare software used by hospitals worldwide. Google says earlier frontier models had missed it.
Gemini 4 Argon: Google's pre-release process
Google says safely releasing a model at this level requires a phased approach. It is taking part in the US government's voluntary process for pre-release model access while gradually expanding availability. It is also using the early cohort's feedback to refine safeguards before a public launch.
The safeguards span four areas, according to the company:
- Misuse: Defences against cyber and CBRN misuse, including monitoring of the model's internal activations, tested by internal and external red teams.
- Prompt injection: Google calls Argon its most resilient model yet against indirect prompt injection attacks.
- Misalignment: Monitoring of the model's chain-of-thought and actions, with execution stopped when necessary.
- Hardened environments: Sealed, isolated sandboxes for high-risk training and evaluations.
How it compares with GPT-6 Astra
Based on Google's own benchmark table, as reported by VentureBeat, Argon leads outright on 12 of 18 benchmarks and ties for first on one. OpenAI's GPT-6 Astra leads outright on three and ties on one. Anthropic's Claude Opus 5.5 leads on two.
Argon's clearest wins are in enterprise tasks. On Harvey's Legal Agent Benchmark it scores 19.6 percent against Astra's 5.4 percent. On Vals Finance Agent v2 it posts 65.4 percent to Astra's 53.5 percent. Astra holds its ground in technical work: it leads Argon by 10.5 percentage points on both FrontierSWE v2 and Terminal-Bench Science 0.1. The two models tie at 68 percent on CWE-bench v1.
Pricing is another contrast. Argon launches at $2 per million input tokens and $10 per million output tokens, one-fifth of Astra's listed $10 and $50. After the introductory period it rises to $4 and $20. On prompt-injection resistance, Google's chart shows Argon at a 0.7 percent attack success rate against 8.5 percent for Astra.
