LLM API cost calculator
Estimate your monthly spend for any model and token volume — then see equal-or-better models that cost less. Prices refresh daily across every major provider and host.
At a glance
Free LLM API cost calculator. Estimate your monthly spend for any model by entering your input and output token volumes, using provider list prices per 1M tokens updated daily. The calculator then surfaces models with an equal-or-better coding-benchmark score that cost less for the same usage, so you can cut spend without cutting quality.
Enter volumes in millions of tokens (e.g. 10 = 10,000,000). Uses provider list prices per 1M tokens; cached-input discounts and tiered rates aren’t applied.
Estimated monthly cost
$110
gpt-5.5 · OpenAI · 80.6% coding
Equal-or-better coding, lower cost
claude-opus-4-7
Anthropic · 83.5% coding
$100/mo
save $10.00 (9%)
google/gemini-2.5-pro
DeepInfra · 83.1% coding
$32.50/mo
save $77.50 (70%)
Prices updated 7 October 2026. See the full LLM API Pricing Index. Benchmarks: coding via Aider polyglot & SWE-bench Verified; reasoning & math via Epoch AI.
How the calculator works
- How is the LLM cost calculated?
- Monthly cost = (input price per 1M tokens × your input millions) + (output price per 1M tokens × your output millions), using provider list prices refreshed daily. Cached-input discounts, batch rates, and long-context surcharges are not applied, so it is a conservative list-price estimate.
- How do I find a cheaper model without losing quality?
- Pick your model, volume, and use case. For coding — the axis we benchmark — the calculator lists models with an equal-or-better coding-benchmark score that cost less, ranked by score. For other use cases (reasoning, vision, tool use, long context) it lists cheaper models that are capable of that task, ranked by saving. Switching to a cheaper host of the same open-weight model is always a zero-quality-risk saving.
- Are these real, current prices?
- Yes — prices come from the StackSpend LLM Pricing Index, synced daily from public sources across OpenAI, Anthropic, Google, xAI, DeepSeek, Mistral, and inference hosts like Groq, Together AI, Fireworks, and DeepInfra.
- Where do the benchmark scores come from?
- Coding comes from the Aider polyglot leaderboard and Epoch AI’s SWE-bench Verified; reasoning from GPQA Diamond; math from MATH Level 5 and Mock AIME — the Epoch AI Benchmarking Hub (epoch.ai), used under CC BY. Vision, tool-use, and long-context are capability filters (not yet quality-scored).