Gemini 4 Argon Costs Half as Much as Claude Opus 5.5 Until November 30, and Then the Prices Match

Price per million tokens

As of publication, Google's pricing page lists Argon on the Gemini API at $2 per million input tokens and $10 per million output tokens through November 30, 2026, with output including thinking tokens. Starting December 1, the price becomes $4 and $20. Argon isn't available on the Gemini API free tier. Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens on Anthropic's API, and Anthropic's pricing page shows no scheduled change.

Caching works differently on the two models. Argon supports implicit caching only, and cached input costs $0.10 per million tokens until November 30 and $0.20 after that. Google doesn't list a separate charge for writing to the cache. Opus 5.5 uses prompt caching, where a cache read costs $0.20 per million tokens, a five-minute cache write costs $5 and a one-hour cache write costs $8.

Both companies offer half-price batch processing. Argon's Batch and Flex tiers cost $1 and $5 per million tokens until November 30 and $2 and $10 after that. Opus 5.5 on the Batch API costs $2 and $10. For faster service, Google sells a Priority tier for Argon at $3.60 and $18 until November 30 and $7.20 and $36 from December 1, while Anthropic offers fast mode for Opus 5.5 at $8 and $40 per million tokens.

Context window and output

Google's model page gives Argon a unified context window of 1,048,576 tokens, shared between input and output. The default synchronous maximum output is 65,536 tokens, and long decode lets a single response run up to the full 1,048,576. Anthropic's model page gives Opus 5.5 a 1M-token context window and a maximum output of 128K tokens, or 300K tokens on the Message Batches API with a beta header.

Argon accepts text, image, video, audio and PDF input and returns text. Opus 5.5 accepts text and images and returns text. Google lists thinking levels of low, medium and high for Argon, while Anthropic says adaptive thinking is always on for Opus 5.5 and is controlled with the effort parameter, which defaults to medium.

Where each model is available

Google opened Argon to everyone on October 13. Developers can use it on the paid Gemini API in Google AI Studio, Google AI Ultra and Pro subscribers can use it in the Gemini app and in AI Mode in Search, and Google Antigravity offers it to Ultra subscribers and, since October 16 Beijing time, to paid Google AI Pro subscribers on a separate weekly Argon quota. Enterprises can reach it through Gemini for business and the API.

Anthropic released Opus 5.5 on September 22 and says it's available on all platforms, including Amazon Web Services, Google Cloud and Microsoft Azure. The model page lists the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS. Anthropic also says fast mode for Opus 5.5 is available in Claude Code, and Anthropic's pricing page includes Claude Code in the Pro and Max plans.

What the official benchmarks show

Each company has published its own benchmark results, and the two sets don't line up neatly. Anthropic's table came out on September 22, before Argon was announced, so it doesn't include Argon. Google's model card for Argon does include a column for Claude Opus 5.5, and Google's methodology page says the Opus 5.5 figures were taken from published reports or leaderboards.

In Google's own figures, Argon scores 77.9% on DeepSWE v1.1 against 74.2% for Opus 5.5, 68.9% against 67.0% on the Vals Index, and 51.3% against 42.5% on AutomationBench. Google's table also shows Argon ahead on Harvey's Legal Agent Benchmark, LVBench and long-context GraphWalks. The same table shows Opus 5.5 ahead on Terminal-Bench 4.0, at 66.4% against 57.4%, on Terminal-Bench Science 0.1, on PostTrainBench and on FrontierSWE v2.

In Anthropic's own figures, Opus 5.5 scores 66.4% on Terminal-Bench 4.0, matching the number in Google's table, and 67.7% on Humanity's Last Exam with tools. Anthropic lists Opus 5.5 at 40.0% on AutomationBench, which differs from the 42.5% in Google's table, and notes that Zapier ran and reported that result without fallback models. Anthropic also says that at this level of capability, benchmark margins have become a less reliable guide to real-world differences.

Which one costs less

Until November 30, Argon's list price is half of Opus 5.5's for standard and batch requests. From December 1, the standard and batch list prices are the same. After that, the bill depends on how each workload uses caching, output length and thinking, and neither company publishes a cost comparison between the two models.