Get started
YOUR MODELS. ONE EVALUATION API.

Know what
your model
can really do.

Add a connection. Run meaningful benchmarks. Measure performance with useful scores and task metadata—not retained execution traces.

Start evaluating Read the docs
Bring your own endpoint · No SDK required
YOUR STACK, NOT A NEW STACK.
Completions APIs
Decision models
Custom endpoints
Plain HTTP
FROM ENDPOINT TO EVIDENCE

The hard part is building the model.
Evaluating it shouldn’t be.

A focused workflow for teams that want to understand their models—not maintain evaluation infrastructure.

01

Connect it once.

Add your API URL, credentials, and any custom headers. Give your connection a name you’ll actually remember.

endpoint: "production-v2"
02

Ask better questions.

Submit a benchmark and sample size through the API. Your run keeps a record of the protocol and configuration it used.

benchmark: "mmlu-test"
03

See the whole picture.

Follow progress, compare task scores, and inspect safe failure classifications—without stored prompts or model outputs.

GET /v1/evaluations/:id
A WORKSPACE THAT KEEPS UP

One home for
every evaluation.

From your first test to your next release, keep connections, evaluation history, and API access in one considered workspace.

Product preview with illustrative sample data.Explore the sample workspace
DEVELOPER-FIRST, NOT DEVELOPER-ONLY

Three fields.
That’s the starting line.

Start evaluations through the API. Use the dashboard to follow their status, inspect results, and manage your connections.

Retry safelyIdempotency keys protect against duplicate runs.
Keep access in your controlScoped, revocable keys for your workspace.
Know what you measuredBenchmark versions and protocol details travel with results.
Read the quickstart
Plain HTTP. No ceremony.
curl "$ALLEVALS_URL/v1/evaluations" \
  -H "Authorization: Bearer $ALLEVALS_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: release-v2-mmlu-001" \
  -d '{
    "endpoint": "my-model",
    "benchmark": "mmlu-test",
    "limit": 25,
    "expected_collection_mode": "prepaid"
  }'
GOOD EVIDENCE STARTS WITH GOOD QUESTIONS

Benchmarks, with
the fine print included.

Explore the catalog

Choose from the benchmarks available to run. Each includes source attribution and scoring details.

LESS GUESSING. BETTER MODELS.

Your next release
deserves a clearer picture.

Create your workspace Start with the docs
No credit card needed to create an account.