Argon vs Opus: Which API Bill Fits the Job?
Gemini 4 Argon lists $2 and $10 per million tokens through November 30, 2026, then $4 and $20. Claude Opus 5.5 lists $4 and $20. On December 1 the batch rates both sit at $2 and $10.
Argon vs Opus is a choice between two API bills that meet on December 1, 2026, and a difference in how long one reply can run. Gemini 4 Argon is gemini-4-argon. Claude Opus 5.5 is claude-opus-5-5, released September 22, 2026. The prices below are the published lists as of October 21, 2026. This page did not call either model.
The list price, before and after December 1
On the Gemini Developer API, Argon's standard price through November 30, 2026 is $2 per million input tokens and $10 per million output tokens. Output includes thinking tokens. From December 1 the standard price is $4 and $20. Argon is not on the free tier.
Opus 5.5's standard price is $4 per million input tokens and $20 per million output tokens. That is the December Argon list, already.
| Gemini 4 Argon, Gemini API | Claude Opus 5.5 | |
|---|---|---|
| Standard input / output | $2 / $10 through Nov 30, 2026. $4 / $20 from Dec 1 | $4 / $20 |
| Batch | Half the standard rate: $1 / $5 through Nov 30, then $2 / $10 | 50% off input and output: $2 / $10 |
| Priority or fast | Priority: $3.60 / $18 through Nov 30, then $7.20 / $36 | Fast mode: $8 / $40. Fast mode is not available with the Batch API |
| Cached input | Implicit cache only. Prompts of at least 4,096 tokens. A hit is $0.10 through Nov 30, then $0.20 | Cache read $0.20. A 5-minute cache write is $5. A 1-hour cache write is $8. The shortest cacheable prompt is 512 tokens |
Through November 30, Argon batch is the lower batch bill. From December 1, the standard rates match and the batch rates match. Argon's Flex tier publishes the same token prices as its Batch tier.
Claude Sonnet 5.5 already publishes $2 / $10 as its standard price, with a 1M context window and a 128K output cap. That rate does not expire on December 1. It is a different model from Opus.
Cache is a different product on each side
Argon does not accept an explicit cache. Calls that create or pass cachedContents are off this model. The 95% discount is implicit, and it applies when the prompt is at least 4,096 tokens. temperature, top_p, and top_k are ignored if you send them.
Opus charges for the cache write, then for the read. A hit is 5% of the base input price, $0.20 per million tokens. Fast-mode prices stack on top of the cache multipliers. Fast mode is a Claude API price. The pricing page also lists it for Claude in Microsoft Foundry when the deployment is hosted on Azure, and leaves it off Claude Platform on AWS and partner-operated clouds.
On Google Cloud Agent Platform, the global standard row matches the Gemini API standard numbers: $2 / $10 through November 30, then $4 / $20. A non-global region is 10% higher. The Cloud page describes that through-November rate as a credits-back promotion, and it says the promotion does not apply to Gemini 4 Argon Cyber Permissive. Cyber Permissive is listed at the December rate. The Gemini API page describes the same window as an introductory list price. Those are two billing notes.
The Agent Platform page also says that from November 13, 2026, a completion longer than 32K tokens is charged for the retrievable tokens even when the request fails on a client or server timeout.
How long one reply can be
Argon's input, thinking, and output share one window of 1,048,576 tokens. A single response can run up to 1M tokens minus the input. Past 64K tokens, the request needs background execution and max_output_tokens above 65,536. If background is on and that limit is unset, the server stops at 64K. max_output_tokens counts thinking and the answer together. The thinking level is low, medium, or high. The default is medium.
Opus 5.5 has a 1M-token context window and a 128K max output on a normal request. On the Message Batches API it can go to 300K output tokens with the output-300k-2026-03-24 beta header. Adaptive thinking stays on. Depth is the effort setting, and the default effort is medium.
A million tokens is not the same amount of text on both tokenizers. Anthropic says Claude 4.7 and later models produce about 30% more tokens for the same text than earlier Claude models, and that the exact increase depends on the text. The Argon docs do not publish a matching word count.
Which bill to open
- You are paying standard rates before December 1. Argon's list is $2 / $10. Opus's list is $4 / $20. Argon is not on the free tier.
- The job can wait in a batch. Through November 30, Argon batch is $1 / $5 and Opus batch is $2 / $10. From December 1, both batch rates are $2 / $10, and both standard rates are $4 / $20.
- One reply has to run past 128K tokens. Argon's long decode uses the rest of the shared 1M window, and it needs background execution. Opus stops at 128K on a normal call, and at 300K on the batch beta.
- You already keep an explicit Claude cache. Argon will not take that cache object. Its discount is the implicit hit on prompts of at least 4,096 tokens.
Gemini 4 Argon and Claude Opus 5.5 are both on Seailife. A separate page compares GPT-6 and Claude 5.5 API prices.