Gemini 4 Argon: Google's New Frontier Model Launches at $2 Per Million — What Buyers Should Know

Google's Gemini 4 Argon arrives at an introductory $2 per million input tokens — the same launch price as GPT-6.1 Sol. But the steady-state price tells a different story, and the 1M-token output ceiling may matter more than the price tag.

Key takeaways

• Gemini 4 Argon (announced Sept 30): Google's new frontier model, built for long-horizon work — software engineering, legal/finance knowledge work, cybersecurity defense.
• Introductory pricing: $2/M input, $10/M output — rising later to $4/$20. Cached input 95% off.
• Headline capability: 1M-token maximum output (up from 64K) — the largest output ceiling Google has shipped.
• Restricted rollout: trusted cyber defenders first (Fairwind Program), then paid API customers and AI Ultra subscribers. No firm public date.
• Bottom line: don't switch on launch price alone. Test only where a measurable bottleneck exists; keep model dependencies behind an abstraction layer.

Google announced Gemini 4 Argon on September 30 — its new frontier model, which Google describes as its most powerful yet. Argon is built to "sustain deep reasoning across complex, long-horizon workflows," with three stated targets: real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense. (Google; TechCrunch)

Two details matter for enterprise buyers. First, Argon raises its maximum output from 64K to 1M tokens — the largest output ceiling Google has shipped. Second, the introductory price: $2 per million input tokens and $10 per million output tokens, later rising to $4/$20.

GPT-6.1 Sol launched at the same $2/$10 price point. The launch-week pricing looks aggressive — but the fine print is where the business analysis starts. (Geeky Gadgets)

What Google actually shipped — and who gets it first

Argon is beginning with a restricted rollout. Rather than a public launch, Google is releasing the model to a set of trusted cyber defenders through its Fairwind Program. Broader access starts with paid API customers and Google AI Ultra subscribers before expanding to developers, enterprises, and consumers.

There is no firm public release date. Google says it is engaged in the US government's voluntary pre-release model access process while it iterates on safeguards. (Google)

The sequencing is deliberate. Argon can autonomously find, validate, and patch software vulnerabilities — capabilities Google describes as dual-use, which is why early access is restricted and safeguards are being strengthened against cyber misuse. For the initial cohort, trusted defenders and Google's own teams receive a version released without cyber guardrails, so they can use its full defensive capability.

Wiz is already using Argon through its Scan for Good program. Google says the model uncovered a critical vulnerability exposing sensitive personal information in healthcare software used by hospitals — a flaw previous frontier models had missed. (Google)

Benchmarks (vendor-reported — validate before buying)

• DeepSWE v1.1 (long-horizon software engineering): state of the art at 77.9%
• Vals Index (finance, coding, legal, tax): top spot
• AutomationBench (end-to-end business workflows): #1 at 51.3%
• LVBench (long video understanding): 91.7%
• CWE-bench v1 (vulnerability remediation): ties for first at 68%

These are vendor-reported results. Useful for comparison — but enterprises should validate on their own workloads before treating them as purchasing evidence.

Why the 1M-token output limit matters

Previous Gemini models capped output at 64,000 tokens. Argon expands that to one million tokens in a single response — which Google calls industry-leading. Output length rarely makes headlines, but for some long-running workflows, it may be more operationally significant than another benchmark gain.

High-value enterprise workloads require long, multi-step execution: the software-engineering and professional knowledge-work tasks Google highlights. Today, long workflows are orchestrated across multiple model calls — the application manages state and intermediate outputs. A higher output limit keeps substantially more generation inside a single model trajectory.

One necessary qualification: a 1M-token maximum is a capacity limit, not a guarantee — not proof that million-token generations will be reliable, economical, low-latency, or desirable on real workloads. Enterprises still need to test those properties themselves.

Google's own examples make it concrete

• Fleet optimization: Argon agents analyzed profiling telemetry and applied memory optimizations across Google's data centers — 300 TiB freed so far, an estimated 500 TiB to 1 PiB in total.
• Language migration: agents migrating C/C++ codebases to Rust at scale — up to the Fuchsia Zircon kernel at 800K+ lines.
• Decoder rewrite: for the libgav1 video decoder, agents replaced 32K lines of SIMD code with safe Rust the compiler could auto-vectorize — a memory-safe decoder running 2.7x faster than the existing Rust port, identical video output.
• Quantum research: Argon beat a published baseline by 40% on algorithmic optimization in minutes.

(Google)

Whether those numbers hold outside Google's walls is unverified. But the pattern is the thing: long, autonomous, multi-step work with more of the reasoning kept inside a single trajectory.

Argon at $2/$10 — before moving to $4/$20

(Google; DX Today; CoinCentral)

Once the introductory period ends, Argon's steady-state price matches Anthropic's cheaper option — not the dramatic undercut the launch-week headlines suggest. Build long-term budgets on the steady-state price, not the intro rate.

Two caveats: these are introductory prices on a model with a staged rollout — Argon isn't broadly available yet. And token prices are only part of the economics: what matters is total cost per successfully completed workflow, not sticker price per token.

What this means for your business

Price is only one differentiator. When two vendors launch at the same price, the decision shifts from "which model can we afford" to "which capability actually removes a bottleneck in our pipeline." Audit your automation workloads for measurable bottlenecks — the tasks where long-horizon reasoning or long outputs are the constraint.

Design longer trajectories with guardrails, not just capacity. A higher output limit lets more of a long task stay in one trajectory — but concentrates risk in a single run. If one agent performs more steps before returning control, you need approval boundaries, checkpointing, resumability, execution logs, and rollback strategies. Present these as design requirements from the start. The model doesn't provide them for you.

Evaluate the defensive tooling, not the headlines. Google reports Argon found a critical vulnerability previous frontier models missed, and Wiz is already using it to protect public infrastructure. Security teams should evaluate whether stronger AI-assisted vulnerability discovery warrants changes to existing processes. That's the defensible takeaway — not that anyone's patch cycle is suddenly obsolete.

Don't switch on launch price alone. A practical decision rule: never migrate because of an introductory price or a benchmark table. Test only if the workload has a measurable bottleneck the new capability could remove. Keep model-specific dependencies behind an abstraction layer. Build against the capability, not the brand.

Measure caching on repeatable workloads. Google's 95% cached-input discount is real, but it matters most for workflows that repeatedly reuse eligible context. At the introductory rate, it would take eligible cached input from $2 to $0.10 per million tokens. Before projecting savings, validate cache-hit rates and eligible context on your actual traffic. (Google)

The bottom line

For every production workflow, measure input tokens per completed task, output tokens per completed task, cache-hit rate, retries, failed runs, and human-review time. Lower token prices only help if they translate into lower cost per successfully completed task.

The practical question isn't which launch generated the strongest benchmark table — it's whether a new model can lower the measured cost or complexity of a specific production workflow.

Want to know which of your workflows are actually ready for long-horizon AI agents — and what each one costs per completed task? Get a free AI audit and we'll map it out with you.

FAQ

What is Gemini 4 Argon?

Google's new frontier model (announced Sept 30, 2026), built for long-horizon work: software engineering, legal/finance knowledge work, and cybersecurity defense. Headline specs: 1M-token maximum output (up from 64K) and introductory pricing of $2/M input, $10/M output.

When will Argon be publicly available?

No firm date. Rollout is staged: trusted cyber defenders first (Fairwind Program), then paid API customers and Google AI Ultra subscribers, then developers, enterprises, and consumers. Google is also going through the US government's voluntary pre-release model access process.

Is Argon really cheaper than competitors?

At launch, yes — $2/$10 matches GPT-6.1 Sol. But the introductory rate rises to $4/$20 later, matching Anthropic's cheaper tier. Headlines about undercutting describe the intro window, not the steady state. Budget on the $4/$20 figure.

What's the most important capability for businesses?

Arguably the 1M-token output ceiling: it lets long multi-step workflows stay inside a single model trajectory instead of being orchestrated across many calls. But capacity isn't reliability — test long generations on your own workloads, and design guardrails (checkpoints, approval boundaries, rollback) from the start.

Sources

• blog.google — Introducing Gemini 4 Argon
• techcrunch.com — Google Releases Gemini 4 Argon, Called Its Most Powerful Model Yet
• coincentral.com — Google's New AI Model Just Undercut OpenAI and Anthropic on Price
• news.dxtoday.com — DX Today AI, October 1, 2026