Gemini 4 Argon is out — Google prices it to undercut OpenAI and Anthropic

Google released Gemini 4 Argon on October 13, ending a nearly year-long gap in its flagship AI lineup. The model is now live for Pro and Ultra subscribers in the Gemini app and AI mode in Search, and for Ultra subscribers on the Antigravity agentic development platform, according to The Verge.

The release follows a rocky stretch for Google's frontier model roadmap. The company last shipped a flagship with the Gemini 3 series in November 2025. A promised Gemini 3.5 Pro update never materialized, and Google spent the summer shipping smaller Flash models instead. New DeepMind chief Koray Kavukcuoglu acknowledged in late September that the company "took a little bit of a step back," but said Gemini 4 was in its refinement stage and would launch "much earlier" than the end of the year, as The Verge reported.

Pricing is the most concrete differentiator. Argon launches at an introductory rate of $2 per million input tokens and $10 per million output tokens — roughly a fifth of the cost of OpenAI's GPT-6 Astra and Anthropic's Fable 5 models. Those prices double at the end of November, but even then Argon will undercut its frontier rivals on a per-token basis. As The Verge notes, that doesn't guarantee lower real-world costs, since models can consume different numbers of tokens for equivalent tasks.

The model also supports a 1-million-token output limit, up from 64,000 in previous Gemini models, which Google says lets users complete longer tasks in a single pass, per Ars Technica.

On benchmarks, Google claims Argon outperforms GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 across coding, cybersecurity, science, finance, and law. On the software engineering DeepSWE v1.1 benchmark, Argon scores 77.9 percent, ahead of those rivals, Ars Technica reports. The benchmarks remain vendor claims; independent verification will have to wait for wider access.

Access is still limited by design. The public rollout comes after a September 30 announcement that restricted the model to a "set of trusted cyber defenders," with Google saying it was "actively engaged in the U.S. government's voluntary process for pre-release model access," as The Verge reported. It is unclear when or if Argon will reach Plus subscribers or free users — those tiers were effectively downgraded earlier this month, when free Gemini accounts were limited to Flash Lite and AI Plus users lost Pro access on October 9, per The Verge.

Google says it is already using Argon internally: patching cybersecurity vulnerabilities, halving the compute power needed for YouTube's comments section, and debugging Alphabet's Wing drone physics simulator. Ars Technica adds that Argon agents migrated C/C++ codebases to Rust across Google, including more than 800,000 lines in the Fuchsia OS Zircon kernel, and that fleet-wide telemetry analysis saved 300 TiB of memory in Google data centers.

The launch lands a day after OpenAI's DevDay, where OpenAI announced its Dots agent and GPT-6.1 Sol, and said it was shelving GPT-6.1 Astra. Google also previewed a "universal" enterprise Gemini agent on October 8 that runs across Workspace, third-party apps like Slack and Microsoft 365, and devices, with a companion identity that can send email from an "@agents.company.com" address. It's available only in private preview to enterprise customers, The Verge reports.

What remains unverified is whether Argon's benchmark performance holds up in real workloads — and whether Google's aggressive token pricing survives once competitors respond. The company has not said when API access widens beyond current tiers.