API

AI IQ publishes a small JSON API for public model, benchmark, ranking, chart, and methodology metadata. The API is designed for simple read-only integrations and returns sanitized public fields only.

Configured clients should use https://www.aiiq.org/api/v1/ as their API base URL. The listed GET endpoints return JSON. AI IQ also exposes a remote MCP connector for agents at https://www.aiiq.org/api/mcp.

The response examples below are abridged for readability. Scores, ranks, timestamps, and model coverage change as source data is refreshed; the field tables and OpenAPI document define the current response shape.

Endpoints

curl https://www.aiiq.org/api/models
curl https://www.aiiq.org/api/models/gpt-5.5
curl https://www.aiiq.org/api/benchmarks
curl https://www.aiiq.org/api/rankings
curl https://www.aiiq.org/api/charts
curl https://www.aiiq.org/api/methodology

MCP Connector

Agents and MCP clients can connect to AI IQ's hosted MCP server with this remote server URL:

https://www.aiiq.org/api/mcp

Use that URL in clients that support remote MCP or Claude custom connectors. In Claude, add a custom connector and paste the URL above as the remote MCP server URL. In other agent clients, choose the remote or Streamable HTTP MCP option and use the same URL.

Name: AI IQ
URL:  https://www.aiiq.org/api/mcp
Type: Remote MCP / Streamable HTTP

The connector exposes nine read-only tools for listing models, fetching model details, listing and fetching rankings, listing benchmarks, listing and fetching domains, reading methodology metadata, and comparing models.

/api/mcp is a Streamable HTTP MCP transport endpoint. It is discoverable after a client is explicitly configured with the URL, but it is not a normal web page or JSON API route to open directly in a browser.

Models

GET /api/models returns a list of public model summaries. Every response is wrapped in an envelope with apiVersion, methodologyVersion, updatedAt, and a models array.

{
  "apiVersion": "1",
  "methodologyVersion": "2026-07-24-arc-coverage-stability",
  "updatedAt": "2026-06-08T00:00:00.000Z",
  "models": [
    {
      "id": "gpt-5.5",
      "name": "gpt-5.5",
      "provider": "OpenAI",
      "rank": 3,
      "iq": 132,
      "dimensions": {
        "abstract-reasoning": 114,
        "mathematical-reasoning": 141,
        "academic-reasoning": 140,
        "programmatic-reasoning": 139,
        "computer-use": 135,
        "reliability": 123
      },
      "emotionalReasoning": 118,
      "cost": {
        "inputPer1M": 5,
        "outputPer1M": 30,
        "ioPer1M": 35,
        "usageMultiplier": 0.974,
        "effectivePer1M": 34.0775,
        "unit": "USD per 1M I/O Tokens"
      },
      "updatedAt": "2026-06-08T00:00:00.000Z",
      "url": "https://www.aiiq.org/models/gpt-5.5/"
    }
  ]
}

Benchmarks

GET /api/benchmarks returns metadata for the 34 benchmarks assigned to the six scored IQ dimensions. Domain benchmarks, including Emotional Reasoning, are exposed through GET /api/domains instead. Benchmark scores and rankings are exposed through GET /api/rankings.

{
  "apiVersion": "1",
  "methodologyVersion": "2026-07-24-arc-coverage-stability",
  "updatedAt": "2026-06-08T00:00:00.000Z",
  "benchmarks": [
    {
      "id": "arc-agi-2",
      "name": "ARC-AGI-2",
      "description": "",
      "category": "iq",
      "dimension": "abstract-reasoning",
      "direction": "higher_is_better",
      "unit": "percent",
      "updatedAt": "2026-06-08T00:00:00.000Z",
      "url": "https://www.aiiq.org/charts/"
    }
  ]
}

Model Detail

GET /api/models/:id returns a single public model with the full dimensions block, cost block, all benchmark results, and the methodology version. The envelope is the same as other endpoints (apiVersion, methodologyVersion, updatedAt) with model fields at the top level.

The dimensions block here is the full form (vs. the compact slug→iq map on /api/models): each entry includes iq, benchmarksCovered, benchmarksTotal, and imputed. Each benchmarkResults entry includes dimension and direction in addition to value, field, name, and category.

{
  "apiVersion": "1",
  "methodologyVersion": "2026-07-24-arc-coverage-stability",
  "updatedAt": "2026-06-08T00:00:00.000Z",
  "id": "gpt-5.5",
  "name": "gpt-5.5",
  "provider": "OpenAI",
  "rank": 3,
  "iq": 132,
  "dimensions": {
    "abstract-reasoning": {
      "iq": 114,
      "benchmarksCovered": 4,
      "benchmarksTotal": 4,
      "imputed": false
    },
    "mathematical-reasoning": {
      "iq": 141,
      "benchmarksCovered": 4,
      "benchmarksTotal": 5,
      "imputed": false
    },
    "academic-reasoning": {
      "iq": 140,
      "benchmarksCovered": 5,
      "benchmarksTotal": 6,
      "imputed": false
    },
    "programmatic-reasoning": {
      "iq": 139,
      "benchmarksCovered": 5,
      "benchmarksTotal": 6,
      "imputed": false
    },
    "computer-use": {
      "iq": 135,
      "benchmarksCovered": 6,
      "benchmarksTotal": 6,
      "imputed": false
    },
    "reliability": {
      "iq": 123,
      "benchmarksCovered": 6,
      "benchmarksTotal": 7,
      "imputed": false
    }
  },
  "emotionalReasoning": 118,
  "cost": {
    "inputPer1M": 5,
    "outputPer1M": 30,
    "ioPer1M": 35,
    "usageMultiplier": 0.974,
    "effectivePer1M": 34.0775,
    "unit": "USD per 1M I/O Tokens"
  },
  "url": "https://www.aiiq.org/models/gpt-5.5/",
  "benchmarkResults": {
    "arc-agi-2": {
      "value": 85,
      "field": "arcAgi2",
      "name": "ARC-AGI-2",
      "category": "iq",
      "dimension": "abstract-reasoning",
      "direction": "higher_is_better"
    }
  }
}

Rankings

GET /api/rankings returns the ranking catalog: id, name, type, direction, model count, and the detail URL for every leaderboard — without the ranked model rows. Fetch one full leaderboard with GET /api/rankings/:id. Use rankingType to distinguish types. The order is: composite-iq (derived), effective-cost (derived), six dimension IQ rankings (type dimension), then per-benchmark rankings (type benchmark). Composite IQ is higher-is-better; Effective Cost is lower-is-better. Emotional Reasoning (EQ) is available through the Domains API rather than as an IQ dimension ranking.

{
  "apiVersion": "1",
  "methodologyVersion": "2026-07-24-arc-coverage-stability",
  "updatedAt": "2026-06-08T00:00:00.000Z",
  "rankings": [
    {
      "id": "composite-iq",
      "rankingName": "Composite IQ",
      "rankingType": "derived",
      "dimension": null,
      "direction": "higher_is_better",
      "modelCount": 97,
      "url": "https://www.aiiq.org/api/rankings/composite-iq"
    },
    {
      "id": "abstract-reasoning-iq",
      "rankingName": "Programmatic Reasoning IQ",
      "rankingType": "dimension",
      "dimension": "abstract-reasoning",
      "direction": "higher_is_better",
      "modelCount": 104,
      "url": "https://www.aiiq.org/api/rankings/abstract-reasoning-iq"
    }
  ]
}

Each detail response carries the standard envelope plus the full ranking (the pre-split inline shape):

{
  "apiVersion": "1",
  "methodologyVersion": "2026-07-24-arc-coverage-stability",
  "updatedAt": "2026-06-08T00:00:00.000Z",
  "id": "composite-iq",
  "rankingName": "Composite IQ",
  "rankingType": "derived",
  "direction": "higher_is_better",
  "models": [
    { "id": "gpt-5.5", "name": "gpt-5.5", "provider": "OpenAI", "rank": 3, "score": 132 }
  ]
}

Charts

GET /api/charts returns promoted canonical chart metadata. Chart URLs are stable share pages, not rendered SVG exports. More / Experimental charts are excluded.

Methodology

GET /api/methodology exposes public methodology metadata and the hierarchy behind the two derived rankings (Composite IQ and Effective Cost). For the full scoring explanation, use the Methodology page.

The Composite EQ standalone ranking was retired. Emotional Reasoning (EQ) is a separate domain built from EQ-Bench 3 and AttuneBench and remains excluded from Composite IQ.

Each scored dimension in the Composite IQ derivedRankings entry carries a slug field that matches the dimension slugs used in model detail and dimension rankings. The six dimensions are Abstract, Mathematical, Academic, Programmatic, Computer Use, and Reliability.

{
  "apiVersion": "1",
  "version": "2026-07-24-arc-coverage-stability",
  "summary": "Composite IQ uses six equally weighted capability dimensions mapped through IQ 70-160 expected-score ladders and conservative imputation rules. Abstract Reasoning equally averages the available source-backed ARC projections; coverage count communicates confidence, while zero-coverage models retain conservative fallback estimates.",
  "derivedRankings": [
    {
      "id": "composite-iq",
      "rankingName": "Composite IQ",
      "direction": "higher_is_better",
      "scoreField": "iq",
      "dimensions": [
        {
          "id": "D1",
          "slug": "abstract-reasoning",
          "name": "Abstract Reasoning",
          "minBenchmarks": 1,
          "benchmarks": [
            { "field": "arcAgi3", "name": "ARC-AGI-3" },
            { "field": "arcAgi2", "name": "ARC-AGI-2" },
            { "field": "arcAgi1", "name": "ARC-AGI-1" }
          ]
        }
      ]
    },
    {
      "id": "effective-cost",
      "rankingName": "Effective Cost",
      "direction": "lower_is_better",
      "scoreField": "effectiveCost",
      "unit": "USD per 1M I/O Tokens",
      "breakdown": [
        {
          "id": "published-pricing",
          "name": "Published Token Pricing",
          "inputs": [
            { "field": "inP", "name": "Input token price" },
            { "field": "outP", "name": "Output token price" }
          ],
          "summary": "Base cost is input price plus output price for 1M input tokens and 1M output tokens. Free, zero-dollar, or rate-limited rows remain null unless a specific nonzero hosted-provider price is recorded."
        },
        {
          "id": "usage-multiplier-waterfall",
          "name": "Usage Multiplier Waterfall",
          "steps": [
            "measured benchmark multiplier from validated token usage and price-adjusted task-cost residuals",
            "one-generation-back same-family same-lineage multiplier",
            "two-generations-back same-family same-lineage multiplier",
            "geometric average of the three closest measured peers",
            "assumed 1x fallback for positive-price models"
          ]
        }
      ]
    }
  ],
  "updatedAt": "2026-06-08T00:00:00.000Z",
  "url": "https://www.aiiq.org/methodology/"
}

Domains

GET /api/domains lists applied-capability domains tracked outside Composite IQ: cybersecurity, bio, machine-learning, finance, writing, software-engineering, webdev-design, emotional-reasoning, legal, and healthcare. GET /api/domains/:slug returns one domain with per-model composite scores (models — the same domain IQ the domain pages display) and its benchmark catalog with results. Domain results are leaderboard rows rather than one-value-per-model maps: each row carries model, harness (the agent scaffold used for the run, when published), and score, so a model can appear once per published harness. Domain benchmarks and scores are not part of Composite IQ or the dimension rankings.

{
  "apiVersion": "1",
  "methodologyVersion": "2026-07-24-arc-coverage-stability",
  "id": "cybersecurity",
  "name": "Cybersecurity",
  "url": "https://www.aiiq.org/cybersecurity/",
  "models": [
    { "model": "gpt-5.5", "compositeIQ": 147, "benchmarksCovered": 4, "benchmarksTotal": 5, "modelUrl": "https://www.aiiq.org/models/gpt-5.5/" }
  ],
  "benchmarks": [
    {
      "id": "exploitbench",
      "name": "ExploitBench",
      "direction": "higher_is_better",
      "unit": "percent",
      "source": { "name": "ExploitBench leaderboard", "url": "https://exploitbench.ai/" },
      "results": [
        { "model": "mythos-5", "harness": "AutoNudge", "score": 78 }
      ]
    }
  ]
}

Compatibility & Versioning

Configured clients should use https://www.aiiq.org/api/v1/ as their canonical API base URL. This is a permanent alias; the bare https://www.aiiq.org/api/ also works and always serves the current version.

The API follows a compatibility policy:

• 2026-07-21 — BREAKING: Composite IQ moved to six equally weighted dimensions: Abstract, Mathematical, Academic, Programmatic, Computer Use, and Reliability. Emotional Reasoning became a separate domain; Software Engineering and WebDev & Design were added as domains. The public benchmark API now exposes only the 34 scored dimension benchmarks.
• Additive changes (new fields, new endpoints, new benchmarks in a dimension) can happen at any time without notice.
• No removals or renames: Fields are never removed or renamed under apiVersion: "1".
• Breaking changes are batched, dated in the changelog below, and — once external consumers exist — will never ship in place. They will arrive at a new /api/v2/ endpoint alongside /api/v1/ with a published sunset window for the bare alias.
• Version numbering: apiVersion uses integer majors only; subversions are never used.

Changelog

Newest first. Changes marked BREAKING require client updates when upgrading.

• 2026-07-24: Added the finance domain with Finance Agent v2, CorpFin v2, Excel Modeling, τ³-Banking, MortgageTax, and TaxEval v2.
• 2026-07-24 — BREAKING: ARC-AGI-1/2/3 projections were jointly recalibrated and source-backed Abstract IQ now equally averages only the direct ARC projections available for the model. Missing ARC estimates remain the fallback for zero-coverage models, while later benchmark reveals are order-invariant. methodologyVersion bumped to 2026-07-24-arc-coverage-stability.
• 2026-07-24 — BREAKING: Abstract Reasoning now uses ARC-AGI-3, ARC-AGI-2, and ARC-AGI-1. Chess Puzzles was removed from AI IQ. ARC-AGI-1 was recalibrated to the official 64.2% average-human anchor. methodologyVersion bumped to 2026-07-24-abstract-reasoning-calibration; the public benchmark API now exposes 33 scored benchmarks.
• 2026-07-10: Added domain composite scores (models array on GET /api/domains/:slug) and the machine-learning domain.
• 2026-07-10 — BREAKING: GET /api/rankings is now a catalog (id, rankingName, rankingType, dimension, direction, modelCount, url) without inline models arrays. Full leaderboards moved to GET /api/rankings/:id. MCP get_ranking resolves through the new endpoint.
• 2026-07-10: Added /api/domains and /api/domains/:slug (cybersecurity, bio); JSON discovery index at /api root; OpenAPI 3.1 spec at /api/openapi.json; /api/v1/ canonical alias; CORS (Access-Control-Allow-Origin: *) and explicit Cache-Control on all endpoints; uniform JSON 404s with a code field; per-benchmark chart URLs in /api/benchmarks. Human docs moved from /api/ to /docs/api/.
• 2026-07-09 — BREAKING: D4/D5 dimension slugs renamed from app-building and production-engineering to frontend-engineering and backend-engineering across all responses. The old slugs no longer appear in dimensions maps, derivedRankings, or benchmark dimension fields. Page URLs 301-redirect; JSON keys do not. methodologyVersion bumped to 2026-07-09-gso-ml-domain (GSO-Bench moved out of Composite IQ scoring into the ML domain).

Privacy Boundary

The public API exposes only sanitized public data intended for read-only integrations. It omits private configuration, unpublished records, internal review metadata, and operational tooling details.