Gemini 3.8 Flash API pricing and cost calculator

How this is calculated

Take a call with 100,000 input and 5,000 output tokens. Uncached, the input costs $0.15 and the output costs $0.0375, so the call comes to $0.1875. The calculator starts with half the input read from cache, which brings the input part to $0.0825 and the call to $0.12, or $12 across 100 calls in a month. That's $0.0825 for input and $0.0375 for output. Which part dominates depends on your mix, so measure your own split before you decide between a cap on response length and a stable cached prefix. While the introductory discount runs, the same half cached call costs $0.06 and 100 of them cost $6. Google lists that discount through December 31, 2026, and calls after that date bill at the standard $12. Three workloads look very different from that starting point. A chat product sends 2,000 input and 400 output tokens per message, and most of that input is a repeated system prompt. A million messages with no cache hits cost $6,000, and with most of the input cached they cost $3,840. An agent session of 20 turns, each carrying 60,000 input tokens that are mostly cached and producing 1,500 output tokens, costs $0.567 for the whole session. A document pipeline that reads 5,000 files of 40,000 input and 2,000 output tokens each costs $375 with nothing cached. Run those three through the discounted rates and the cached chat case becomes $1,920, the agent session $0.2835 and the document run $187.50. Treat those as temporary figures.

Verdict

Don't choose Flash on rates alone. Run your own prompts through it, record input tokens, cached tokens, output tokens and retries, and price the finished results rather than single calls. Anything that can wait until morning belongs in the batch queue, where input sits 50% below the standard rate and output sits 50% below it. Count image, video and audio inputs with your own samples, since they turn into tokens you can't guess from file size. The token counter on this site helps with the text part. A single response that fills the 65,536 token output limit costs $0.49152 in output alone, which is the ceiling to hold against your max token setting. Before you commit, rerun the budget at standard rates, because the discount stops after December 31, 2026.