Gemini 4 Argon is gated: budget at $4/$20, not the intro price
Google's Gemini 4 Argon tops coding benchmarks but ships to cyber defenders first. Its $2/$10 intro price doubles later. Plan your routing for the real rate.
Google released Gemini 4 Argon today and called it its most powerful model yet. You cannot use it yet. Gemini 4 Argon goes first to a short list of cyber defenders. Paid API customers come "as soon as possible," with no date. And the launch price is an introductory price that doubles later. If you plan to move agent work to Argon, do the math on the second number.
What actually happened
Per Google's announcement, Gemini 4 Argon targets software engineering, legal and finance knowledge work, and cybersecurity defense. The facts that matter for a buyer:
- Access: limited to "trusted cyber defenders" through Google's Fairwind Program. Paid API customers and Google AI Ultra subscribers are next. No date.
- Intro price: $2 per million input tokens, $10 per million output. Cached input gets a 95% discount.
- Later price: $4 input, $20 output after the introductory period. Google did not say how long that period lasts.
- Output limit: up to 1 million output tokens per response, up from 64K.
- Google's benchmarks: 77.9% on DeepSWE v1.1, 51.3% on AutomationBench, 68% on CWE-bench v1 (vulnerability fixes, tied for first).
TechCrunch reports Google's claim that Argon beats GPT-6 Astra and Anthropic's Fable and Opus models across several benchmarks. Those are vendor numbers. Nobody outside Google's partner program can check them on real work yet.
Why the Gemini 4 Argon price matters for your business
The intro price is a hook. At $2/$10, Argon matches GPT-6.1 Sol, which OpenAI shipped yesterday and which you can call today. At $4/$20, Argon costs twice as much per token. Build your cost model on the number you will pay in month six, not the number on launch day.
Three moves for this week:
Do not rewrite your routing for a model you cannot call. A benchmark lead on a gated model is a press release. Keep shipping on what is generally available.
Write the eval now. Pick 20 real tasks from your own agents: a support reply, a code fix, an invoice extraction. Score your current model on them today. When Argon opens up, you run the same set and get an answer in an afternoon.
Watch the 1M output limit. Long output is useful for codebase migrations and big documents. It is also a way for one runaway agent to spend $20 on one response at full price. Put a max_output_tokens cap on every call before you switch.
Key takeaways
- Gemini 4 Argon launched September 30, but only for cyber defenders in Google's Fairwind Program
- Paid API access comes next, with no date given
- Intro price is $2/$10 per million tokens; it rises to $4/$20 after an unstated period
- Benchmark wins are Google's own claims and are not yet testable by most buyers
- Build a 20-task eval on your own work now so you can test Argon the day it opens
A new model every week is only useful if switching costs you one afternoon. We build model-agnostic AI systems with eval sets, routing and spend caps you own, so a launch like Argon is a config change, not a rebuild. See how we build it, or run your AI cost numbers.
Sources: Google, TechCrunch.
- #gemini-4-argon
- #api-pricing
- #model-routing
- #ai-costs
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
White House AI accord is morally binding. Audit your vendors
Six AI leaders signed a White House AI accord on internal controls and outside auditors. It has no legal teeth. What your vendor contracts should say instead.
Read itOpenClaw Enterprise: a free agent control plane, pilot only
OpenClaw Enterprise is an MIT-licensed, self-hosted control plane for persistent AI agents from OpenAI, Red Hat and Nvidia. It is free, and it is pre-1.0.
Read it