Artificial Analysis API

• Create a key in API key management. Keys belong to organisations, so pick the one with the subscription.
• Call the API with the key in the x-api-key header. Keep it server-side.

curl "https://artificialanalysis.ai/api/v2/language/models" \
  -H "x-api-key: YOUR_API_KEY"

Responses share one envelope: tier, pagination on lists, and data. A 401 means a bad key; a 403 means your tier doesn't cover the endpoint.

Using an AI assistant instead? Connect the MCP server.

Send your API key in the x-api-key header. There's no OAuth or request signing for the API. The MCP server uses OAuth sign-in, and takes the same key for deployed agents.

Keys belong to organisations. Regenerating a key in API key management revokes the old one immediately. Keep keys server-side and out of source control.

Your plan decides which endpoints you can call. Others return 403.

Quotas run on a fixed 24-hour window that starts with your first request and resets in full when it ends. Every key in the same scope (one user in an organization, or the whole organization) shares one quota. Signed-in MCP gets double the usage. An MCP call made with an API key counts as one request.

Responses carry X-RateLimit-Limit, X-RateLimit-Remaining, and X-RateLimit-Reset. A 429 adds Retry-After in seconds. Contact the team for more.

Every authenticated response carries a few headers that describe the request context and quota state.

These conventions apply across the API.

• Dates use ISO 8601: 2025-06-15 for calendar dates and 2025-06-15T14:30:00.000Z for timestamps.
• Country codes use lowercased ISO 3166-1 alpha-2 values: us, cn, gb.
• End-to-end response times assume 500 answer tokens unless a field name states otherwise.
• Field names use snake_case. We keep the wire format stable: fields are deprecated, not renamed.
• Nulls mean "not measured" or "not applicable". They never mean zero.

artificial_analysis_intelligence_index is a composite score, and what goes into it evolves. We version the index so you can interpret a score in context.

Current version: v4.3.2. Full composition and version history on the Intelligence Index methodology page.

How versions change:

• Major version bumps indicate a substantial change to the index definition or interpretation. Scores are best interpreted within the same major version.
• Minor version bumps indicate methodology changes within the same major version. These may include changes to contributing evaluations, task subsets, graders, scoring/normalization, weightings, anchors, or evaluation refreshes. Scores may or may not change.
• Patch version bumps indicate lower-significance methodology changes of the same kinds — for example upgrading an evaluation's grader or dataset version. Scores may change. They are not reflected in intelligence_index_version, which reports major.minor only.

Currently included (v4.3.2, 10 evaluations): AA-Briefcase v1.1, GDPval-AA v2.1, AutomationBench-AA (partial score), Terminal-Bench 4.0, SciCode, AA-LCR v1.1, AA-Omniscience, HLE, GDP.pdf, CritPt.

artificial_analysis_coding_index and artificial_analysis_agentic_index are derived from subsets of these evaluations and are not separately versioned.

The six Capability Indexes (artificial_analysis_finance_and_accounting_index, artificial_analysis_strategy_and_ops_index, artificial_analysis_legal_index, artificial_analysis_healthcare_and_medical_index, artificial_analysis_engineering_index, artificial_analysis_economics_index) are versioned separately from the Intelligence Index. Each is null for models missing any evaluation it depends on. Composition on the Capability Indices methodology page.

Use of the API requires attribution across all tiers. When you display or share API data, credit Artificial Analysis as the source. A visible byline or footer link is sufficient.

Use of the API is also subject to our Terms of Use and Data Platform Terms. For redistribution rights or bespoke contract terms, contact the team.

Errors return a JSON body with an error string and, when useful, a details object. The HTTP status identifies the failure category.