Gemini 4 Argon vs Claude Opus 5.5: API pricing, caching, and batch

How they differ

Standard input costs $4 per million tokens on Argon and $4 on Opus. Output costs $20 and $20, and cache reads cost $0.20 and $0.20. A request with 100,000 input and 5,000 output tokens costs $0.50 and $0.50 with no cache hits, and $0.31 and $0.31 with half the input read from cache. While Google's introductory discount runs through November 30, 2026, Argon bills $2 input, $10 output and $0.10 for cache reads. The half cached example then costs $0.155 on Argon during the discount, against $0.31 at standard rates and $0.31 on Opus. Standard rates start December 1, and Google's page says the cache is implicit only from that date. Budget a product that will still be running in December on the standard rates, since the discount only covers the early weeks. Argon has Batch API prices here, so a nightly job of 20,000 documents with 6,000 input and 800 output tokens each and no cache hits costs $400 through batch, against $800 at standard rates. Opus has Batch API prices too, which bring the same job to $400. On the Claude side a cache write costs $5 per million tokens for five minutes, and $8 for an hour. Opus lists no separate long prompt rate here. Argon has no separate long prompt tier. Both models list a window near a million tokens, so a 400,000 token document costs $1.68 on Argon and $1.68 on Opus at the rates that apply. These examples exclude cache writes and tool fees. The embedded chart follows live OpenRouter rates, which can depend on provider routing.

Verdict

Price both on a standard footing first, then add what applies to you. While the introductory discount runs, Argon can look better than it will later, and that's the number most people copy into a spreadsheet. Use the standard rates for any plan that runs past November 30, 2026, and treat the discount as a short window to test on. Then look at the cache. Opus charges a write fee when a prefix enters the cache, so an agent that keeps rewriting a changing prefix pays for it. Google's page says Argon's cache is implicit only from December 1, so check how your setup reuses prefixes before then. Then look at batch. Both have Batch API prices here, so deferred work gets a discount on either side. Finally, add tool fees. Google Search grounding on Argon is free for 5,000 prompts a month, shared across Gemini 3 and later models, then costs $14 per 1,000 prompts. A product that grounds 50,000 prompts a month pays for 45,000 of them, which is $630. Price any search or tool use on the Claude side from Anthropic's page before you compare totals.