Google moved Gemini 4 Argon out of its cybersecurity-only preview and into the paid Gemini API on October 13, pricing it at $2 per million input tokens and $10 per million output tokens through November 30. On December 1 both numbers double, to $4 and $20. The model is also live in AI Studio, Antigravity, eligible Gemini Enterprise subscriptions, and the Gemini app's AI Pro and AI Ultra tiers.

Until September 30 Argon was restricted to vetted defenders inside Google's Fairwind Program. Six weeks later it's a routing decision for anyone with a billing account.

The pricing structure is the real signal. Google is subsidizing a seven-week window in which production teams can run Argon against incumbents at roughly half the eventual sticker, which is exactly the behavior Google wants: dependency built during the discount, re-priced at full rate once workflows have hardened.

The case for burning cycles on that bake-off sits in the benchmarks. On Zapier's AutomationBench, which measures multi-step business-process completion, Argon posts 51.3% against Claude Opus 5.5 at 42.5% and GPT-6 Astra at 41.4%. On the Vals AI GDP-weighted index of professional work, Argon ranked first on October 12 at 68.90%, ahead of Claude Sonnet 5.5 at 67.04%. The picture is less clean elsewhere. On the Artificial Analysis Intelligence Index, Argon scores 53, tied with GPT-6 Astra and Claude Fable 5.1 and trailing Claude Opus 5.5 at 58, Claude Sonnet 5.5 at 56, and Meta Muse Wonder Full at 56. On Terminal-Bench 4.0, Opus 5.5 leads at 66.4% versus Argon's 57.4%.

Translation: Argon is the frontier on long-horizon business execution and professional-work reasoning. It isn't the frontier on raw intelligence aggregates or on terminal agentic coding. The benchmark Argon dominates is also the one that most closely proxies the kind of chained task a five-to-thirty-person team actually runs through an API, research, draft, revise, hand off.

The enterprise reference list Google surfaced through Reuters, carried via the Galveston County Daily, reads accordingly: Deloitte, LexisNexis, Ambience Healthcare, Isomorphic Labs. Three of those four are regulated-knowledge-work shops. Google is positioning Argon less as a chat model than as the throughput tier for workflows that previously snapped on step four.

The DeepMind leadership shakeup referenced in Reuters' coverage sits in the background of the launch rather than in its text. What the pricing cliff suggests is a company trying to convert benchmark narrative into billed tokens before the next model cycle, Anthropic's or OpenAI's, resets the comparison table. The 2023 GPT-4 API rollout followed the same template: aggressive introductory pricing, a short window, and durable workloads parked on the platform by the time the discount ended.

After November 30, the question stops being whether Argon wins AutomationBench and starts being whether the gap justifies paying twice as much per token as it did the week before.

Gemini 4 Argon Hits the Paid API at $2/$10 — Introductory Rate Expires November 30

Sources