Gemini 4 Argon Pricing & Access: What Changed

Google announced Gemini 4 Argon on September 30, 2026, with an introductory token rate and frontier claims for coding, enterprise work, and cybersecurity. The crucial limitation: Argon is not broadly available to developers, enterprises, or consumers yet.

The live table, chart, and calculator above use Google’s announced rate card from the pricing dataset. They are planning tools, not evidence that a public gemini-4-argon API route is callable today.

What Google announced

Argon is rolling out first to trusted cyber defenders through Google’s Fairwind Program. Google says it is participating in the U.S. government’s voluntary pre-release access process and will expand to paid API customers and Google AI Ultra subscribers after additional feedback and safeguard work. It gave no public rollout date.

Google also raised the maximum output claim from 64K to 1 million tokens. That is an output ceiling, not a published input-context window. Google has not yet supplied a public API model ID, detailed model card, rate limits, or stable endpoint documentation.

Gemini 4 Argon pricing impact

Google’s introductory rate is $2 per 1M input tokens, $0.10 per 1M cached input tokens, and $10 per 1M output tokens. The announced rates match GPT-6.1 Sol across fresh input, cached input, and output. Argon’s 95% cache discount could matter for repeated policies, repository maps, document sets, and tool instructions, but Google has not published cache-storage charges or long-context tiers specifically for Argon.

After the introductory period, Google’s footnote says the price becomes $4 input and $20 output per 1M tokens. Google gives no end date. If the stated 95% cache discount continues, cached input would calculate to $0.20 per 1M; that future cache figure is arithmetic, not a separately quoted rate.

Do not treat the introductory rate as an available production SKU. The public Gemini pricing catalog does not yet list an Argon endpoint, and access through Fairwind is a restricted program. Use the token calculator to model future traffic, then confirm the final API terms before procurement.

What the benchmarks show—and do not show

Google reports 77.9% on DeepSWE v1.1, first place on the Vals Index, 51.3% on AutomationBench, 91.7% on LVBench, and a joint-leading 68% on CWE-bench v1. It also describes internal deployments that optimized quantum subroutines, freed more than 300 TiB of data-center memory, and helped migrate large C/C++ codebases to Rust.

These are vendor-reported results and case studies. Google has not published enough raw traces, prompts, complete cost ledgers, or repeated-run evidence here to reproduce every claim.

Artificial Analysis has now tested Gemini 4 Argon at High reasoning. It reports a 53 Intelligence Index score, matching GPT-6 Astra at Max and one point above GPT-6.1 Sol at Max. The independent run cost $1.99 per Intelligence Index task during the introductory discount and averaged roughly 62,000 output tokens per task; Artificial Analysis projects $3.98 per task at the future list price. That is a benchmark-specific result, not a universal application cost or an AI Pricing Guru Labs score.

The same analysis reports 78% on AutomationBench-AA, 57% on Terminal Bench 4, a 15% hallucination rate among its evaluated leading models, and a one-million-token context window. We retain the context figure as third-party evidence only: Google’s announcement explicitly confirms a one-million-token output limit but does not publish the input-context specification.

Who benefits—and who should wait

Potential early winners: approved cyber defenders, vulnerability-remediation teams, and Google partners evaluating long-horizon coding or professional workflows under controlled access.

Most developers should wait: there is no broadly callable API route, rollout date, public input-context specification, or independent speed result. Keep high-volume work on an available Flash model; compare OpenAI pricing and Anthropic pricing when immediate frontier access matters.

Security teams need extra controls: Google’s trusted defenders may receive Argon without cyber guardrails, while broader releases will use misuse protections. Fairwind access should not be interpreted as permission to run unscoped offensive tests.

What developers should do now

1. Do not invent an API ID or silently route production aliases to Argon.
2. Build a frozen evaluation set for coding, knowledge work, prompt injection, and authorized vulnerability remediation.
3. Record accepted results, output length, latency, cache hits, retries, reviewer time, and safety refusals.
4. Reprice the workload when Google publishes the final endpoint, context limit, cache-storage terms, and availability date.
5. Keep a non-Google control. Check Novita’s current model catalog for an OpenAI-compatible comparison route.

Affiliate disclosure: AI Pricing Guru may earn a commission from the sponsored Novita link at no extra cost to you. It does not affect this analysis.

Bottom line

Gemini 4 Argon is an announced frontier model with unusually large output headroom and an aggressive introductory rate card. It is not yet a public API launch. Buyers should separate Google’s vendor benchmarks and future pricing from today’s restricted access, then evaluate the model only when the exact route and terms are available.

Sources checked October 1, 2026: Google’s official Gemini 4 Argon announcement, the Fairwind Program, the independent Artificial Analysis model page and benchmark analysis, plus the launch discussion and pricing discussion on Hacker News. For current production choices, read the Gemini API pricing guide and Google AI pricing page.