What This Calculator Measures
Every AI provider prices its models differently, and the gap between the cheapest and most expensive option for the same task can be 50x or more. This calculator takes one prompt and expected output length, and prices it across every current OpenAI, Anthropic, and Google Gemini model side-by-side, so you can see exactly where the money goes.
How the Comparison Works
The same prompt is tokenized once per tokenizer family (exactly for OpenAI's o200k_base models, as an estimate for Claude/Gemini since they don't publish an offline tokenizer), then priced using each model's real, currently published input and output rate per million tokens. Your expected output token count is applied against every model's own output price.
Why Prices Vary So Much
- Model size/capability tier. Flagship "reasoning" models (like GPT-5.6 Sol or Claude Opus 5) cost far more per token than their smaller, faster siblings (GPT-5.6 Luna, Claude Haiku 4.5) because they're larger models doing more computation per token.
- Long-context surcharges. Some models charge a higher rate once a request crosses a length threshold (for example, several current models roughly double their per-token price past 200K-272K input tokens).
- Provider positioning. Providers price competitively against each other in similar capability tiers, but the exact numbers shift often as new model generations ship.
How to Use This for Real Decisions
- Paste the actual prompt structure you'll send in production (including system prompt/instructions).
- Set the expected output length to match your real completions, not a guess — this often matters more than the input side.
- Sort by cost and look at the gap between the cheapest model that still meets your quality bar and the one you were about to default to.
Common Mistakes
- Comparing sticker price without comparing quality. The cheapest model per token isn't cheapest per successful task if it needs retries or produces worse output.
- Ignoring caching. If your workload repeats a large system prompt or document across calls, the cheaper model on paper might not stay cheaper once you factor in cache discounts on a different model — see the Prompt Caching Savings Calculator.
Frequently Asked Questions
Frequently Asked Questions
Every current model we track across OpenAI, Anthropic, and Google Gemini, using each provider's published input/output pricing.
Yes for OpenAI models (exact, offline tokenizer). For Claude and Gemini, the same prompt is tokenized with an estimate, clearly labeled, since no public tokenizer exists for those models.
Yes — set the expected output tokens field and every model's cost updates using its own output price per million tokens.
Related Calculators
You might also find these tools helpful