API documentation
The Modelglass read API serves the same pricing + ontology data as the comparison site, with full pricing history and the structured feed for operators. Paid tiers require an API key.
Get an API key →View the machine-readable OpenAPI spec → /v1/openapi.json
Response fields
Responses contain exactly the fields documented in the OpenAPI spec, which marks
each data object additionalProperties: false. Open-ended
records (capabilities, tier
attributes) are the only exceptions. New fields are
added to the spec when they're served, without a schema_version
bump, so tolerate fields you don't recognise. Every notes
field is public commentary written for you (a sale price, a regional caveat); it may
be absent, and provenance always lives in source.
Base URL
https://modelglass-api.vercel.appAuthentication
Send your API key as a Bearer token on every /v1
request. Missing key → 401; over your plan limit
→ 429. Successful responses carry
X-RateLimit-Limit/Remaining/Reset headers.
Authorization: Bearer mg_starter_devkeyEndpoints
| Method | Path | Description |
|---|---|---|
| GET | /v1/meta | Artifact version, built_at, content hash, counts |
| GET | /v1/models | All models; filter ?provider= ?unit= ?status= ?arch= |
| GET | /v1/models/:modelId | One model by model_id (e.g. bfl/flux-1-schnell) |
| GET | /v1/models/:modelId/competitors | Closest competitors with current_price and price_delta_ratio; a retired competitor has current_price null and a last_price instead |
| GET | /v1/providers | All providers |
| GET | /v1/providers/:slug | One provider by slug |
| GET | /v1/benchmarks | All benchmark definitions |
Pricing units
Every price row carries a unit string. Current values:
per_imageper_megapixelper_secondper_stepper_clipper_1m_tokens_inputper_1m_tokens_outputper_1m_tokensper_1k_tokens_inputper_1k_tokens_outputper_requestper_monthper_creditper_1m_charactersper_minuteper_generationper_1k_requestsper_1m_tokens_citationper_1m_tokens_reasoningper_1m_tokens_cache_readper_1m_tokens_cache_writeper_gpu_second
unit is documented as a string, not a closed list: new units are added
without a version bump, so treat a unit you don't recognise as "not comparable", not as an error.
Prompt caching. per_1m_tokens_cache_read and per_1m_tokens_cache_write are prompt-cache
rates (cached-input reads and cache writes), on their own tiers next to input and output. They're real
prices for cache-heavy workloads, but never a model's headline price.
Headline rows. The MCP pricing feed
(modelglass_get_pricing_feed) returns one row per active tier, each with a
headline boolean. Rows with headline: false
(Batch, Flex, off-peak and cache rates) are real but not the list price. Filter on
headline: true before taking the cheapest row per unit.
Plans & rate limits
| Plan | Pricing history | Rate limit |
|---|---|---|
| Free | Active price only | 100 req/day |
| Starter | Full pricing history | 10,000 req/day |
| Pro | Full history | 100,000 req/day + SLA |
Free tier reads the static artifact (no key needed) · browse all model pricing →
Examples
# Metadata
curl -H "Authorization: Bearer mg_starter_devkey" \
https://modelglass-api.vercel.app/v1/meta
# All models, filtered to one provider
curl -H "Authorization: Bearer mg_starter_devkey" \
"https://modelglass-api.vercel.app/v1/models?provider=replicate"
# One model (full pricing history on Starter+)
curl -H "Authorization: Bearer mg_starter_devkey" \
https://modelglass-api.vercel.app/v1/models/bfl/flux-1-schnell