Google opened Gemini 4 Argon to the paid Gemini API on October 13, 2026, at an introductory rate of $2 per million input tokens and $10 per million output tokens. On December 1, those rates double to $4 and $20. The window is six weeks, it's explicitly priced as a window, and it lands on the model that currently tops Zapier's AutomationBench.

Argon scored 51.3% on AutomationBench, the Zapier benchmark that measures end-to-end execution of multi-step business workflows. Claude Opus 5.5 scored 42.5%. GPT-6 Astra scored 41.4%. On Harvey's Legal Agent Benchmark the spread is wider: Argon at 19.6%, GPT-6 Astra at 5.4%, Opus 5.5 at 3.8%. Vals AI's independent GDP-weighted index put Argon in first place at 68.90% as of October 12.

The benchmark picture isn't a clean sweep. On Artificial Analysis's Intelligence Index, Argon ties GPT-6 Astra and Claude Fable 5.1 at 53, trailing Claude Opus 5.5 at 58 and both Claude Sonnet 5.5 and Meta's Muse Wonder Full at 56. Opus 5.5 also leads Terminal-Bench 4.0 at 66.4% versus Argon's 57.4%. The honest read is that Argon is the strongest publicly available model for business-process execution, not that it's the strongest at everything.

Then there's price. Opus 5.5 lists at $4/$20 per million tokens. GPT-6 Astra lists at $10/$50. Through November 30, Argon is literally half of Anthropic's rate and one-fifth of OpenAI's, with a cached-input tier at $0.10 per million, a 95% discount on repeat context. The context window is 1,048,576 tokens, shared between input and output, with a 65,536-token default max output.

For a founder or a five-to-thirty-person team weighing whether to build an AI-driven outreach or content pipeline, the forcing function is specific: integrate before December 1 and the unit economics stay cheap; integrate after and they don't. That still assumes someone wants to build the pipeline. LemonLime sells the opposite proposition, a done-for-you service, $999/month for a single business, that studies the company and prepares finished prospect outreach, content, and distribution opportunities proactively, with the owner approving sends and publishes. The model-selection question is Google's problem, not the buyer's.

The broader context is in prior coverage of Argon's AutomationBench debut and the paid-API rollout mechanics. Google has essentially priced a frontier model as a loss leader for forty-eight days. Whoever's doing procurement knows exactly what that means.

Sources

https://blog.google/innovation-and-ai/products/gemini-app/gemini-4-argon-availability/
https://gemini.google/release-notes/
https://venturebeat.com/technology/google-makes-gemini-4-argon-broadly-available-with-limited-time-low-api-pricing-of-2-10-per-1m-token
https://thenewstack.io/gemini-4-argon-live/
https://felloai.com/gemini-4-argon/