Gemini 4 Argon Costs $1.99 a Task at Promo Price Against $0.72 for GPT-6.1 Sol, and Is Not Yet on Sale
A.I. / news
Gemini 4 Argon Costs $1.99 a Task at Promo Price Against $0.72 for GPT-6.1 Sol, and Is Not Yet on Sale
Google's introductory $2 and $10 rates match OpenAI's Sol per token, but Artificial Analysis figures cited by eesel show Argon writing 62,000 output tokens a task where GPT-6 Astra writes 27,000.
Google announced Gemini 4 Argon on Sept. 30 at $2 per million input tokens and $10 per million output tokens, then said the rates double to $4 and $20 after an introductory period it has not dated. Nobody outside Google's Fairwind Program can buy it yet.
The announcement, signed by Koray Kavukcuoglu, senior vice president of Google DeepMind and Google's chief AI architect, says Argon is rolling out first to trusted cyber defenders through Fairwind. Developers on the paid API and Google AI Ultra subscribers come "soon", with no date. The model's output limit rises to 1 million tokens from 64,000, Google said. Cached input tokens are 95 percent cheaper than the input rate.
Google's own benchmark figures
Every score in the post is Google's, and the post does not say any were independently verified. It reports 77.9 percent on DeepSWE v1.1, 51.3 percent on Zapier's AutomationBench, 91.7 percent on the long-video test LVBench and a 68 percent top score, tied for first, on CWE-bench v1. For Vals, Harvey's legal benchmark and Gray Swan's prompt-injection test it says only that Argon leads, without a score.
Google also lists vendor-reported engineering results. Agents moved C and C++ code to Rust, including a libgav1 video decoder that Google says runs 2.7 times faster than the existing Rust port with identical output, and Google says memory optimisations will free more than 300 TiB across its data centres.
What a task costs, per Artificial Analysis
The per-token match is exact. eesel, a support-software vendor that published a pricing breakdown on Oct. 1, lists GPT-6.1 Sol at $2 input and $10 output, identical to Argon's promotional rates, and Claude Opus 5.5 at $4 and $20, identical to Argon's standard rates.
The cost per task is not close. eesel cites Artificial Analysis, which runs a composite intelligence index, as putting Argon at $1.99 per index task at promotional prices and $3.98 at standard prices. GPT-6.1 Sol is $0.72 and GPT-6 Astra is $3.26.
- GPT-6.1 Sol (max)0.72 USD
- Gemini 4 Argon (high), promo1.99 USD
- GPT-6 Astra (max)3.26 USD
- Gemini 4 Argon (high), standard3.98 USD
Source: Artificial Analysis figures as cited by eesel, Oct. 1, 2026
The driver is verbosity. eesel reports Argon used 62,000 output tokens per task against 27,000 for Astra, and produced 110 million tokens over the index run against a median of 82 million. On the same index eesel lists intelligence scores of 54 for Opus 5.5, 53 each for Argon and Astra, and 52 for Sol. eesel puts Argon's cost at about 2.7 times Sol's for scores a point apart.
| Model | Input / output per million tokens | Index score |
|---|---|---|
| GPT-6.1 Sol | $2 / $10 | 52 |
| Gemini 4 Argon, promo | $2 / $10 | 53 |
| Gemini 4 Argon, standard | $4 / $20 | 53 |
| Claude Opus 5.5 | $4 / $20 | 54 |
What is missing
On Oct. 1, eesel says, a generateContent call to the model name gemini-4-argon returned 404 NOT_FOUND, and Argon is absent from Google's API pricing page. Unpublished are long-context surcharges, Batch, Flex and Priority rates, and any free tier. eesel's own worked examples, such as a 1,000-run agent workload moving from $620 to $1,240 a month after the promo, are its calculations and not measurements, and its recommendation to budget at $4 and $20 comes from a company that sells a support A.I. product.
Google's safety section is specific about the delay. It says broad availability waits on stronger safeguards, and that trusted defenders and Google's internal teams get a version without cyber guardrails. The model can find, validate and patch critical vulnerabilities on its own, per the post. For how an agent's candidate findings thin out in practice, see Cloudflare's audit-skill funnel, where 20,799 candidates became 7,245 actionable findings.
For comparison with the other tier of the market, Claude Haiku 5.5 costs $0.10 per million input tokens, and ChatGPT's GPT-6 Sol is already in the Chat tab.
The date to watch is the end of the introductory period, because it is the one figure on which the $1.99 and the $3.98 depend, and Google has not given it.
Sources
More in A.I.
- 01LTX-2.5's Free Commercial Licence Stops at $10 Million in Group Revenue, and Its Card Asks for Contact DetailsLightricks' open-weight video and audio model is free for production use below that line, but the revenue test counts parent companies and affiliates, and the card publishes no benchmark scores.
- 02EmbeddingGemma 2 Embeds Text, Images, Audio and Video in 567MB of RAM on a Pixel 11 ProGoogle DeepMind's Apache 2.0 embedding model has 740M parameters in three modular pieces, and the only benchmark number its launch post prints is a 9.92-point gain on MTEB Code.
- 03Kolibri-1 Is Apache 2.0 and Fits on One B200, but Trails Qwen3.8 27B by 9.1 Points in GermanAleph Alpha's 78B-parameter mixture-of-experts model activates 3.46B parameters per token, and its own model card shows a larger dense Qwen ahead on every headline benchmark.
- 04Qwen3.8-27B Is Apache 2.0 With a 262,144-Token Window and Already Has 482 Finetunes, Including Cloudflare's ClefAlibaba's dense 27B model claims 61.7 on SWE-bench Pro against 53.4 for Opus 4.6 Max, and the model card says nothing about training data.