Google launches Gemini 4 Argon, rejoining the frontier AI race with aggressive pricing

Google has released Gemini 4 Argon, its first major frontier model since Gemini 3 arrived in late 2025. The launch, covered by The Verge, ends a stretch in which Google watched OpenAI and Anthropic ship headline models while it stayed out of the frontier conversation — a period that also included significant internal shakeups and open questions about whether the company still intended to compete at the top end of the market.

Where Argon is available

According to Google's AI blog, Argon is globally available now through the API in AI Studio and the Gemini platform for developers and enterprises, inside the Antigravity agentic development platform for Google AI Ultra subscribers, and in the Gemini app and AI Mode in Search for both Ultra and Pro subscribers. The Verge reports that there is no timeline — and no guarantee — for access on the Plus tier or for free users.

What Google claims

Google positions Argon as a model for deep, multi-step work in coding, finance, law, cybersecurity and science, and claims it outperforms rival frontier models including OpenAI's GPT-6 Astra, Anthropic's Claude Fable 5.1 and Claude Opus 5.5 across a range of benchmarks. The company also says the model ranks first on the Vals Index, which measures performance across economically critical sectors such as finance, coding, legal and tax work. On its product page, Google adds that Argon ties for first on CWE-Bench for vulnerability remediation at 68%, and describes it as its most resilient model against indirect prompt injection attacks.

Beyond benchmark scores, Google emphasises output quality, saying Argon hedges less, repeats itself less and avoids the clichés that still plague chatbot responses.

Customer feedback published by Google is positive but vendor-selected. Box reports a 14% accuracy improvement over Gemini 3.1 Pro in its evaluation; Deloitte says long-running tasks completed successfully 100% of the time in its tests; Kepler says Argon matched analyst-built reference figures on more than 99% of its financial analysis benchmarks; and Cartwheel reports nearly 100x faster runtime and 90% lower token usage in its 3D reconstruction work.

Internal proof points

Google lists several engineering results from its own use of the model:

• A Chrome team used Argon to separate core webpage handling from a roughly six-million-line legacy folder within Chromium's 40-million-line codebase, completing in weeks an architectural refactor expected to take months.
• Working autonomously overnight on the open-source Agent Development Kit, Argon produced optimizations now live in google/adk-python that cut in-memory session preparation time by 40% in long sessions and sped up cold imports by 21%.
• Researchers made the quantum chemistry package xTB up to 2.3x faster at predicting molecular structures and 1.8x faster at conformer search, with the changes contributed upstream.
• Argon delivered a 2.85x inference speedup for the open Gemma models on an A100 via vLLM, raising throughput on Gemma 4 E2B from about 37,900 to over 108,000 tokens per second with no quality loss.
• In Antigravity, Google says agentic code submissions rose by more than 35% while code rollbacks fell by nearly a third.

The Verge adds that Google is already using Argon internally to find and patch security vulnerabilities, to halve the computing power behind YouTube's comment systems, and to debug the drone physics simulator used by Alphabet's delivery unit Wing.

Aggressive pricing

Argon launches at an introductory $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted 95%. According to The Verge, that is roughly a fifth of the price of OpenAI's GPT-6 Astra and Anthropic's Fable 5 models. The rate doubles when the introductory period ends in late November, but Argon should still be significantly cheaper than its rivals per token — although, as The Verge cautions, per-token prices do not map directly to real-world costs, since different models consume different numbers of tokens for the same task.

A staged safety rollout

In line with other recent frontier releases, the public launch followed a limited deployment to a select group of trusted cyber defenders to test for potential risks. Google says it is now expanding its Fairwind security program to hundreds of additional partners, and that Wiz is already using Argon for its Scan for Good initiative, which protects critical infrastructure at no charge.

Why it matters

Google's long absence from the frontier conversation had started to look like a strategic retreat; Argon is a clear signal that it intends to compete. For developers and enterprises, pairing claimed frontier-level benchmark results with pricing far below rival flagships puts immediate pressure on competitors — especially if real-world costs track the token prices. And arriving one day after Meta's Muse Wonder Full, the launch confirms that the frontier race now has at least four serious contenders rather than two.