Gemini 4 Argon API pricing and cost calculator

How this is calculated

The calculator's default request has 100,000 input tokens and 5,000 output tokens. At standard rates it costs $0.50 with no cache hits and $0.31 with half the input cached. Over 100 calls a month the chart starts at $31. With the discount, the half cached call costs $0.155 through November 30, 2026, and $0.31 at standard rates after that. A software engineering agent shows where the money goes. Take 60 turns, each carrying a 120,000 token prompt and writing 4,000 tokens, with most of the prompt read from cache. One session costs $8.976 at standard rates and $4.488 with the discount. If the cache misses, the standard figure becomes $33.60. Until the discount ends, check Google's page for how caching works. From December 1 the cache is implicit only, so Google applies hits on its own and there's no setup step. What you control is the prompt. Keep the stable part first, and read the cached token count in the usage the API returns. Math and science problems flip the balance toward output. A short problem of 8,000 input tokens that produces 40,000 reasoning and answer tokens costs $0.832, and 1,000 of them cost $832. The discounted figure for those 1,000 problems is $416. Batch prices exist too. A nightly job over 10,000 files, each with 25,000 input tokens and 2,000 output tokens, costs $1,400 at the normal rates and $700 in batch mode, which takes 50% off input and 50% off output. Check Google's pricing page for how the discount and batch prices combine before you stack them in a budget.

Verdict

Plan from standard rates and treat any discount as a bonus. The introductory discount runs through November 30, 2026, so a budget built on discounted figures will be wrong from December 1. The standard figures are the safe ones. Argon's output is where reasoning work spends its money, so test it on real engineering, math or agent tasks and log output tokens per task along with accepted results. Search grounding needs its own line in the budget, since the free allowance is shared across Gemini 3 and later models and the fee beyond the allowance is charged per 1,000 prompts. Recheck Google's pricing page before you commit to a volume.