Gemini 4 Argon

Google is back in the frontier race. After months on the sidelines, the company has released Gemini 4 Argon to Pro and Ultra subscribers — claiming frontier-level performance across coding, cybersecurity, science, finance, and law, at roughly a fifth of the token price of its rivals.

What launched

Announced September 30 and rolled out to the Gemini app on October 13, Argon is Google’s first major frontier release since Gemini 3 arrived in late 2025. It’s a sparse mixture-of-experts model trained on TPUs, with native multimodal support — and one headline spec: a 1 million token output limit, up from 64,000 on earlier Gemini releases.

Google’s launch post, written by DeepMind SVP and chief AI architect Koray Kavukcuoglu, claims Argon beats OpenAI’s GPT-6 Astra and Anthropic’s Claude Fable 5.1 and Opus 5.5 on benchmarks including DeepSWE v1.1 (77.9%), LVBench (91.7%), and CWE-bench v1 (68%, tied for first). As always, these are Google’s own numbers — independent verification will follow.

The price war angle

The aggressive part isn’t the benchmarks — it’s the pricing. Introductory API rates of $2 per million input tokens and $10 per million output are about 80% below GPT-6 Astra and Claude Fable. The catch: those rates double at the end of November when the promotion ends.

Even so, the signal is clear. Google is using its TPU cost advantage to pressure the frontier on price, not just capability.

Already working inside Google

Google says Argon is already running internally: patching security vulnerabilities autonomously, halving the compute behind YouTube’s comment moderation, and debugging the physics simulator for Wing’s delivery drones. One striking claim: Argon found a critical vulnerability in healthcare software that previous frontier models had missed.

No word yet on when — or whether — Plus and free-tier users get access. For now, the frontier has a new, cheaper seat at the table.

Sources: Google’s launch post, GenAInews, MarkTechPost