A direct API cost comparison: headline rates, the per-call math on a typical workload, and which one to pick beyond price.
On a standard 1,500-input / 500-output call, Gemini 3.1 Pro is about 55% cheaper — $0.0090 versus $0.0200 per call. At 10,000 calls a month that is roughly $90.00 vs $200.00.
Output tokens cost several times more than input on both models, so the more your workload leans toward long generated answers, the more the output rate dominates the bill.
| Metric | Claude Opus 4.8 | Gemini 3.1 Pro |
|---|---|---|
| Vendor | Anthropic | |
| Input price / M tokens | $5.00 | $2.00 |
| Output price / M tokens | $25.00 | $12.00 |
| Context window | 1M | 1M |
| Cost per typical call | $0.0200 | $0.0090 |
| Cost per 10,000 calls | $200.00 | $90.00 |
Beyond the per-token math, tokenizer efficiency on non-English text and your own quality evals can shift the real cost. Price your actual prompt in the language you serve with the calculator before committing.
Both providers support prompt caching (cached input bills at roughly 10% of the standard rate) and batch processing (about 50% off for 24-hour-tolerant jobs). If one model lets you cache a larger static prefix in your setup, it can become cheaper in practice even when its headline rate is higher.
Gemini 3.1 Pro wins on raw price for a typical workload. Pick Claude Opus 4.8 only when its quality, context, or tokenizer advantages on your specific task outweigh the difference.
Gemini 3.1 Pro — about 55% cheaper on a typical 1,500-input / 500-output call ($0.0090 vs $0.0200).
Claude Opus 4.8: $5.00 input / $25.00 output. Gemini 3.1 Pro: $2.00 input / $12.00 output.
The one with the lower output rate, since generation is output-bound: Gemini 3.1 Pro ($12.00/M output).