Built for people who need to know what to buy for a real AI workload—not merely which advertised unit price is lowest.
Our Mission
AI API Cost exists to answer one practical question: “What is the least expensive way to buy the usable AI output I need?”
Providers bill by tokens, seconds, images, characters, credits, and subscriptions. We normalize those rules to the complete workload, exclude offers that cannot satisfy the request, and show the math behind the recommendation.
What We Compare
- Official APIs, third-party API routes, and subscriptions where cash costs are comparable
- Complete workload cost including retries, billing increments, credits, capacity, and documented free tiers
- Compatibility constraints such as workflow, resolution, duration, audio, latency mode, and model features
- Provider-direct price, billing, capability, and access sources
- Live benchmark scores (Intelligence Index, GPQA, MATH, HumanEval
- Feature matrices for vision, tool use, JSON mode, reasoning, context windows
- Speed measurements (tokens/sec, time-to-first-token)
- Real-world agent performance (PinchBench)
Data Sources
All data comes from public, verifiable sources:
- Provider-direct pages — buying-decision prices, billing rules, capabilities, and access routes
- OpenRouter — third-party LLM route pricing and model metadata
- Artificial Analysis — cross-vendor benchmark and speed data
- PinchBench — real-world agent task scores
We don't make up numbers. Every datapoint is sourced.
Who Built This
AI API Cost is built and maintained by an independent developer. The product stays deliberately narrow: auditable workload decisions that one maintainer can keep accurate.
Corrections
If a price, billing rule, capability, or route is wrong, email us. We investigate sourced-data corrections within 48 hours and preserve the correction in the repository history.
Editorial Standards
Every datapoint on this site is sourced to a public, verifiable source. We do not publish made-up numbers, sponsored placements, or “sponsored bias” in benchmark rankings.
- Sourcing. Each model card, comparison, and benchmark chart links back to its data origin (OpenRouter, the provider's pricing page, or Artificial Analysis for cross-vendor benchmarks).
- Freshness. Pricing is re-fetched hourly from upstream APIs; benchmark scores are refreshed when Artificial Analysis publishes updates. The “Last refreshed” timestamp on each page documents when the source was last verified.
- Corrections. If you spot an error (wrong price, stale benchmark, dead comparison page), email us and we will investigate and fix within 48 hours. Corrections are documented in the GitHub commit history.
- Editorial independence. Affiliate and sponsor placements are kept clearly separate from raw data tables and benchmark rankings. Sponsors never see ranking numbers before the public does.
- AI-assisted prose. The site uses data-driven narrative sections to summarise model positioning. Each summary is generated from the same public data you see above — never from vibes or speculation. We do not publish AI-text without underlying data.
Methodology
For benchmarks we rely on Artificial Analysis as the primary cross-vendor source. Their methodology scores each model on the same set of tests (Intelligence Index, Coding Index, Math Index, GPQA Diamond, AIME, MATH-500, HumanEval variants, etc.) so numbers are comparable across providers.
Buying decisions use provider-direct pricing, billing, capability, and access pages as primary evidence. OpenRouter's public listing powers a separate third-party LLM market explorer and is never allowed to silently override a conflicting provider-direct rate.
For speed we use Artificial Analysis' median tok/s and TTFT measurements, taken on standard hardware and reported per-model.
Contact
Questions, feedback, bug reports, feature requests:
hello@aiapicost.com
← Back to AI API Cost