About APIbenchmarks
APIbenchmarks is a community research initiative. It exists because choosing an API provider for a given workflow is too often decided by marketing and vibes rather than evidence. We collect, score, and publish neutral benchmarks so that engineers, and the AI agents that increasingly make these calls, can make an informed, use-case-specific decision instead of trusting a vendor's landing page.
Who runs it
APIbenchmarks is an open, independent initiative maintained as a community effort by a group of engineers and researchers. We publish benchmarks for the categories developers and agents most often need to evaluate, and we grow the set based on what our users keep asking for. We accept no payment from benchmarked providers, for placement, weighting, ranking, or removal. Contributions and corrections from the community are welcome.
Why neutral, and how we keep it that way
- No paid placement. Ranking is computed, never sold. Providers cannot pay to appear, move up, or be removed.
- One formula for everyone. The same weighted criteria are applied identically to every provider in a category.
- Absolute scales. Scores use fixed reference scales, so a provider's rating does not change when a competitor is added or removed.
- Sourced and dated. Every figure links to a primary source and carries the date it was last verified.
How a benchmark is produced
- 1 · Provider selection. For each category we identify the most widely adopted providers a team would realistically shortlist.
- 2 · Data collection.We record each provider's stated capabilities, pricing, and operational characteristics from their public pricing pages and official documentation, first-party sources only.
- 3 · Scoring. Each provider is scored on four criteria (documentation and developer experience, reliability, ecosystem breadth, and accessibility), combined into the APIbenchmarks Index (ABI) by a published, deterministic formula. Price is reported but kept out of the composite, because price units are not comparable across categories.
- 4 · Verification & dating. Figures are checked against their sources and stamped with a verification date. Anything requiring live, controlled measurement (end-to-end latency, observed uptime) is marked as reserved rather than estimated.
- 5 · Open publication. Aggregate scores are published on the site and through the public JSON API, so the result is auditable and re-computable by anyone.
What we publish openly
| Per-provider scores | / | The full index and category reports |
| REST JSON API | /api/benchmark.json | No auth · CORS open · CC BY 4.0 |
| OpenAPI 3.1 spec | /openapi.json | Machine-readable API contract |
| Agent docs | /agents.md | Guidance for AI assistants |
| LLM site map | /llms.txt | llms.txt convention |
| Methodology | /methodology | Every mapping and weight |
Corrections & contact
If you are a benchmarked provider, or anyone, and you think a figure is wrong, email corrections@apibenchmarks.com with the provider, the specific figure or category, and your evidence (ideally a primary source). We re-verify, correct, and update the verification date, publicly. For anything else, reach us at info@apibenchmarks.com.
Scope & roadmap
Version 1.0 covers 131 APIs across 18 categories, scored from public, first-party sources. A future phase will add live, controlled measurement, latency and observed uptime, under the same open, neutral principles.
Open and free under CC BY 4.0. Last updated 2026-06-27.