LLM token economics
Cache-hit rate, blended $/MTok, and token-type cost shares for a time range. The blended rate uses provider-reported cost. Providers don't report cost per token type, so by_type is priced from configured list prices instead and carries priced_tokens — the share of its tokens that pricing covered. For the users driving that cost, page /analytics/llm-top-users-by-cost.
Cache-hit rate, blended $/MTok, and token-type cost shares for a time range. The blended rate uses provider-reported cost. Providers don't report cost per token type, so by_type is priced from configured list prices instead and carries priced_tokens — the share of its tokens that pricing covered. For the users driving that cost, page /analytics/llm-top-users-by-cost.
ApiKeyAuthAuthorizationBearer <token>Bearer token authentication. Send an API key (a saai_api_-prefixed token) as Authorization: Bearer <token>.
Required permission
view_llm_analyticsstart_time*stringStart of the time range to report on, inclusive.
date-timeend_time*stringEnd of the time range to report on, exclusive.
date-timeprovider?array<>Filter to specific LLM telemetry providers.
model?array<string>Filter to specific model names.
user_id?array<>Filter to specific users by id.
user_group?array<string>Filter to users in specific groups.
auth_mode?array<>The request has succeeded.
application/json- response
Cache economics and token-type cost shares for a time range.
by_type*array<>cache_hit_rate*numbercache_read / (input + cache_read + cache_creation), 0–1.
doublecost_per_mtok*numberBlended reported USD per million tokens.
doublecurl -X GET "https://example.com/analytics/llm-token-economics?start_time=2019-08-24T14%3A15%3A22Z&end_time=2019-08-24T14%3A15%3A22Z"{ "by_type": [ { "token_type": "string", "tokens": 0, "priced_tokens": 0, "weighted_cost": 0.1 } ], "cache_hit_rate": 0.1, "cost_per_mtok": 0.1}LLM spend summary GET
Subscription-vs-API cost split for a time range, using configured seat pricing. Accepts only seat-coherent filters: provider, user, and group scope both the usage legs and the priced seats; a model can't scope a seat, so there is no model param (and no auth_mode — the endpoint decomposes by auth path itself).
LLM token time series GET
LLM token usage over time.