# AI Model Rankings

> Compare large language models on price, context window, capabilities, and real benchmark scores.

Workspace: ai-models. AI Model Rankings tracks 17 published models.

## Workspace

- Directory: https://www.optisift.com/en/workspaces
- Private workbench: https://www.optisift.com/en/w/ai-models/build

## REST API

- Index (no auth): https://www.optisift.com/api/v1?workspace=ai-models
- Health (no auth): https://www.optisift.com/api/v1/health?workspace=ai-models
- Metrics and decision operators (no auth): https://www.optisift.com/api/v1/metrics?workspace=ai-models
- Evidence source registry (no auth): https://www.optisift.com/api/v1/sources
- Match sources to criteria: https://www.optisift.com/api/v1/sources?criterion={criterion}&criterion={criterion}
- Registry responses include runnable connectors, dataset-rights manifests, learning references, and the latest scheduled reachability/shape check.
- OpenAPI 3.1: https://www.optisift.com/openapi.json?workspace=ai-models
- Rankings (auth): GET https://www.optisift.com/api/v1/rankings?workspace=ai-models
- Item (auth): GET https://www.optisift.com/api/v1/items/{slug}?workspace=ai-models
- Compare (auth): GET https://www.optisift.com/api/v1/compare?items={slug},{slug}&workspace=ai-models
- Decide (auth): POST https://www.optisift.com/api/v1/decide?workspace=ai-models
- Decide returns `alternates` (another surviving item per lens, with a note such as close or cheaper) and an `apex-hmac-v2` envelope that signs the decision together with its timestamp and workspace.
- Tracked comparison findings (auth): GET https://www.optisift.com/api/v1/comparisons/{id}/findings?workspace=ai-models

Workspace selection: query `workspace=ai-models` or header `x-apex-workspace: ai-models`.
Auth: `Authorization: Bearer apex_<key>`. The key must belong to the selected workspace, or the key's user must own it. Rate limit: 300 requests/hour per API key.

## MCP

- Endpoint: https://www.optisift.com/api/mcp?workspace=ai-models
- Transport: streamable-http
- Tools: `rank`, `compare`, `explain`, `decide`, `sources`, `plan_dataset`, `preview_source`, `import_source`, `columns`, `saved_comparisons`, `watch_findings`, `ingest`
- Generic client configuration: {"mcpServers":{"apex":{"url":"https://www.optisift.com/api/mcp?workspace=ai-models","headers":{"Authorization":"Bearer apex_<your-key>"}}}}

## Evidence

Start source discovery with the OptiSift evidence registry or MCP `plan_dataset`. It currently exposes 12 review-gated, one-click connectors plus broader discovery and downloadable dataset leads. Educational explainers are returned separately and are never treated as evidence. Use `preview_source` before `import_source`; imports require URL-bound, unchanged schema and rights fingerprints plus explicit owner acknowledgement. Every listed source has a free access path, but dataset-specific rights still require review. Every metric value may include a source, confidence level, verification timestamp, and stable citeUrl. Prefer recent, high-confidence values and treat missing constrained values as failures rather than assumptions.