Google has made Gemini 4 Argon available to customers, ending the run of access limits and configuration hints we have covered since 1 October and 13 October. Google announced the release on the Google Cloud blog, and the arrival comes with a price cut that expires in seven weeks.
The post is specific about where the model lives, and short on labels. As of the announcement, developers can reach Argon in Google Antigravity, Google’s coding environment, and in Google AI Studio. The API in Google’s Gemini platform serves it under the model ID gemini-4-argon. Business users get it inside Gemini for business, in the browser, the desktop app, and on phones. The post never says “general availability” or “preview,” names no regions, and does not mention Vertex AI. Teams that need a data-residency answer or a service-level commitment will have to find it in the model card the post links to.
Pricing is the clearest part. Output tokens cost $10 per million and input tokens $2 per million through 30 November 2026, and cached input is discounted 95 percent. After that date the standard rate doubles to $20 and $4. Anyone budgeting a pilot should plan on the higher figure, since the discount is a seven-week promotion rather than a price.
Google positions Argon for long, multi-step enterprise work where the model has to hold context and apply domain knowledge, and it names software development, finance, legal work, and cybersecurity. The benchmark claims follow that pitch, and every number comes from Google’s post. Google says Argon leads the Vals Index, which scores economic impact across finance, coding, legal, and tax work, and Harvey’s Legal Agent Benchmark. It also reports a tie for first on CWE-Bench v1 with 68 percent on fixing security vulnerabilities, and an attack success rate of 0.7 percent across 15 adaptive attempts on the Gray Swan indirect prompt injection test. Those are third-party benchmarks, but the post gives only Google’s reading of them, with no independent comparison against rival models.
Customer quotes come from partners with a stake in the launch. Harvey’s head of applied research says Argon improved substantially over Gemini 3.1 Pro on its legal agent tests, with the biggest gains in cybersecurity, M&A on the corporate side, and data privacy. ServiceNow reports roughly 20 percent better task success than Gemini 3.8 Flash in early access on its own NowAI-Bench, a test it helped build. Neither is a neutral reading.
Google’s internal examples are the most concrete evidence. Working overnight under automated testing guardrails, Argon found changes to Google’s open-source Agent Development Kit. Google says they trimmed session preparation time held in memory by 40 percent in long sessions and sped up cold imports by 21 percent. Google also says the Chrome team used it to restructure code in hundreds of files of the 40-million-line Chromium project, finishing in weeks what would have taken months.
The security story is where this release connects to our first piece. The general version has safeguards that block offensive work such as penetration testing and exploitation. A version without those guardrails goes only to vetted defenders in Google’s Fairwind Program, which restricts access to cybersecurity firms, government agencies, and operators of critical infrastructure. So the vetted-access model we reported on in October still exists, but now as a tier above the public one rather than the only one.
The practical consequence arrives on 30 November. Teams that start on Argon at $2 and $10 should price their workloads at $4 and $20 before committing, and should run their own tests, because the only comparison on the table is Google’s.
Based on Google Cloud’s blog post of 13 October 2026.
