Skip to content
AgentScoreBETA
Test your docsAPI

The API benchmark
for agents.

Real agents run real tasks on real APIs. We score how far they get and where the friction starts, with the evidence behind every number. So you can pick the API your agent will actually finish the job with.

Explore the benchmarks How we measure

✓ Real executions ✓ Public evidence

ResendEmail API
76.4/100
StripePayments API
77.2/100
TursoDatabase API
65.1/100
ExaWeb search API
80.8/100

Inside the benchmark.

ResendPostmarkSendGridStripeSquarePaddleBraintreeTursoFirebase FirestoreCloudflare D1AppwriteExaFirecrawlTavilyQdrantPineconeWeaviate

Same task.
Different possibilities.

The model is only half the stack. See which APIs help it finish the job, and where the friction starts.

17 scored surfaces across 5 categories. The rest is being measured.
Web search#1 in its category

Exa

REST · exa-rest

80.8/100

AgentScore

78% coverage · measured Oct 5, 2026

Doc readability
57
Error recovery
49
Call efficiency
100
Payments#1 in its category

Stripe

REST · stripe-rest-v1

77.2/100

AgentScore

98% coverage · measured Oct 5, 2026

Doc readability
24
Error recovery
80
Call efficiency
67
Email#1 in its category

Resend

REST · resend-rest-v1

76.4/100

AgentScore

98% coverage · measured Sep 21, 2026

Doc readability
54
Error recovery
42
Call efficiency
100
Vector search#1 in its category

Qdrant

REST · qdrant-cloud-rest

69.5/100

AgentScore

88% coverage · measured Oct 5, 2026

Doc readability
17
Error recovery
36
Call efficiency
100
Database#1 in its category

Turso

REST · turso-http-v2

65.1/100

AgentScore

78% coverage · measured Oct 5, 2026

Doc readability
31
Error recovery
51
Call efficiency
77
All scores out of 100. Higher is better. Unmeasured never means zero.Explore the rubric

One wrong parameter.
Two different futures.

One API points to the problem. Another returns success. For an agent, that response changes everything.

UNKNOWN PARAMETER?agentscore_nonexistent_param=1
stripe-rest-v1HTTP 400
Actionable error

An error.
And a way forward.

Error clarity85/100

The request is rejected. The response gives the caller an error to investigate.

"code": "parameter_unknown"
"message": "Received unknown parameter: agentscore_nonexistent_param"
"doc_url": "https://stripe.com/docs/error-codes/parameter-unknown"
Inspect the full trace
tavily-restHTTP 200
Silent failure

Looks successful.
Hides the mistake.

Error clarity40/100

The unknown parameter goes unflagged. The caller gets no error signal to act on.

{"query":"agentscore","follow_up_questions":null,"answer":null,"images":[],"results":[{"url":"ht…
Inspect the full trace

Actual HTTP responses. Error-handling probes, not a claim of successful task completion. See how the benchmark works

A single number
never tells it all.

Find the trade-offs that matter to your agent. The best choice starts where the average stops.

Interaction Cost51/100
Payload Efficiency5/100

Twilio SendGrid

Fewer calls.
Heavier responses.

An efficient path to sending an email. A very different story when you look at the payload.

Explore the evidence
Offload Value94/100
Safety & Reversibility42/100

Braintree

A useful service.
Unsafe to retry.

Strong service value. But the same request sent twice created two customers: every retry after a timeout becomes a duplicate.

Explore the evidence

Your next user
might be an agent.

Start where they start. See how readable your docs really are.

Free. No account or API key needed. Checks public documentation without JavaScript; this is a partial assessment, not a full API score.

You see the benchmark.
Your agent can use it.

The same evidence. Ready for your agent’s context, your tooling or your next integration.

{ }

REST API

Query the catalogue

Scores, criteria and coverage. Request only what you need.

/api/v1
↳

llms.txt

Give your agent the context

A compact entry point to the catalogue and its methodology.

/llms.txt
⌘

OpenAPI

Build on the benchmark

Endpoint definitions and schemas, ready to integrate.

/openapi.json