Gemini 4 Argon Pricing: What It Costs per Task

Last updated: October 2026

Key takeaways

• Introductory Gemini 4 Argon pricing is reported at $2 input and $10 output per 1M tokens, doubling to $4 and $20 later.
• Cached input at $0.10 per 1M tokens is the lever that decides your real bill.
• Output tokens carry most of the cost, and Argon supports up to 1M output tokens, so cap generation length.
• Access is initially limited to government and trusted cyber-defense partners, so model costs now but do not build a launch plan on it yet.
• Always price a task, not a token, and re-run the forecast when the introductory rate ends.

How much does Gemini 4 Argon cost per million tokens?

According to third-party coverage cited in Google's launch context, introductory pricing is $2 input and $10 output per 1M tokens, rising to $4 and $20 later, with cached input at $0.10. Google DeepMind positions Argon as its most powerful model, with advanced reasoning. Treat the price figures as reported, not as a rate card, and confirm them on the provider page before you commit.

You can see how Google's other models are listed on the Google provider pricing page.

| Tier | Input $/1M | Cached input $/1M | Output $/1M | |---|---|---|---| | Introductory (reported) | $2 | $0.10 | $10 | | Later (reported) | $4 | not reported | $20 |

What does a real workload cost on Gemini 4 Argon?

Say a support bot handles 50k conversations a month. Each conversation sends 3,000 input tokens, of which 2,000 are a cached system prompt and knowledge snippet, and returns 500 output tokens. These numbers are illustrative.

That is 150M input tokens in total: 100M cached and 50M uncached, plus 25M output tokens.

| Line item | Tokens | Rate | Monthly cost (illustrative) | |---|---|---|---| | Cached input | 100M | $0.10 | $10 | | Uncached input | 50M | $2 | $100 | | Output | 25M | $10 | $250 | | Total at intro rate | | | $360 |

At the later reported rate of $4 and $20, and assuming the cached rate stays at $0.10, the same workload costs $200 for uncached input, $500 for output and $10 for cached input, so about $710. That is a cost per conversation of roughly $0.0072 now and $0.0142 later.

The pattern to notice: output is 69% of the introductory bill, and it is the part caching cannot touch. Cheap input does not rescue a chatty model. For the same logic applied to a different launch, see our breakdown of cost per token versus cost per task.

Is Gemini 4 Argon cheaper than GPT-6.1 Sol?

On the headline numbers they look identical at launch: OpenAI lists GPT-6.1 Sol at $2 input, $0.10 cached and $10 output per 1M tokens. The difference is the trajectory. Argon's reported step-up to $4 and $20 doubles your bill, so a fair comparison needs both periods and a quality test on your own prompts. You can line up the two providers on the OpenAI vs Google comparison page.

When should a founder actually use Argon?

Argon is aimed at hard reasoning and long outputs, with access initially limited to government and trusted cyber-defense partners. If your product is a high-volume chatbot, a cheaper tier from the same family is usually the better fit. Reserve a frontier model for the 10 to 20 percent of requests that truly need it, and route the rest. Our post on Gemini 3.6 Flash pricing and agent margins shows what the cheaper end looks like.

What should you model before launch day?

Build the forecast with three scenarios: introductory price, later price, and later price with output capped at half. Add a cache hit rate you can defend, not the best case. Then set your plan price so gross margin holds in the middle scenario. The source for this story is Google's own announcement, Gemini 4 Argon, which is worth reading for the capability claims.

Takeaway: a launch price is a promotion, so price your product on the later rate and treat the introductory rate as upside.

Want to run your own numbers? Try the LLM cost calculator with your own token split.