LLM Provider Price and Latency Monitor

Pricing across 200+ LLMs changes weekly. Tracking it by hand stopped scaling at the 3rd FinOps spreadsheet.

This actor turns public model data into a normalized JSON feed for AI builders, LLM gateways, agent platforms, model marketplaces, and finance teams that need repeatable pricing and capability snapshots. OpenRouter is the primary source because its public models API exposes canonical model IDs, pricing, context length, architecture, supported parameters, and provider routing data. Direct provider pages are used only as limited fallbacks when OpenRouter does not have a model.

The actor runs in Apify Standby mode, so it can be called like a small API. It also supports batch input for scheduled snapshots.

Quick Start

Filter models that support tool calling:

curl "https://api.apify.com/v2/acts/YOUR_ACTOR_ID/runs?token=YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"mode":"models","capability":"tools","provider":"anthropic","limit":5}'

Benchmark a family for chat routing:

$curl "https://YOUR_STANDBY_URL/benchmark?model_family=claude-3&task=chat&limit=2"

Standby Endpoints

Endpoint	Purpose	Billing
GET /	Service info and endpoint list	Free
GET /health	Readiness and service info	Free
GET /models	Filter model snapshots by provider, capability, price, context, and limit	model-snapshot per row
GET /model?id=	Return one full model profile with provider routings	model-snapshot once
GET /benchmark	Compare provider routes for a model family and task	benchmark-row per row
GET /pricing-change	Return current v1 pricing history status	Free in v1
POST /watchlist	Enrich up to 50 requested model IDs	model-snapshot per returned model

All endpoints honor health check values such as ping, test, example.com, and URL forms like https://example.com/status. These short circuit to a mocked LLM shaped response and do not charge.

Model Snapshot Schema

Field	Type	Notes
model_id	string	Canonical OpenRouter slug when available
name	string	Display name
provider_org	string	Source organization such as anthropic, openai, meta, or mistral
providers_serving	string array	Routes serving the model, including OpenRouter and endpoint provider tags
pricing_per_1m_tokens	object	Prompt, completion, image input, and flat request prices
context_length	number	Token context window
max_output_tokens	number or null	Advertised max completion tokens
supports_vision	boolean	Based on model architecture input modalities
supports_tools	boolean	Based on tools or tool_choice support
supports_json_mode	boolean	Based on response_format or structured_outputs support
modality	string	text, multimodal, image, or audio
source	string	Current primary source is openrouter
fetched_at	ISO date	Snapshot time

Benchmark Row Schema

Field	Type	Notes
benchmark_id	string	Stable hash of family, task, model, and providers
model_family	string	Family requested, such as claude-3
task	string	chat, vision, tools, or json
providers	object array	Provider, model ID, token pricing, context, and chat pair cost
cheapest_provider	string	Lowest cost route for a 500 prompt plus 500 completion pair
fastest_advertised_provider	string or null	Uses OpenRouter endpoint latency when present
best_context_provider	string	Route with the largest context window
cheapest_vs_most_expensive_multiplier	number	Most expensive route divided by cheapest route

The total_per_1k_chat_pair_usd metric assumes 500 prompt tokens and 500 completion tokens. It is a simple apples to apples price for one common chat exchange.

Pipeline

flowchart LR
  A[Input or Standby request] --> B[OpenRouter models API]
  A --> C[Limited direct provider fetchers]
  B --> D[Normalize model schema]
  C --> D
  D --> E{Benchmark requested}
  E -->|yes| F[Provider route aggregation]
  E -->|no| G[Model snapshots]
  F --> H[Dataset and JSON response]
  G --> H

Pricing

Event	Price	Charged when
actor-start	$0.50	Once per paid Standby request or batch run
model-snapshot	$0.025	Per enriched model profile returned
benchmark-row	$0.15	Per model family comparison row returned

Charges fire only after real work succeeds. Health checks and empty not found responses are not billed.

Comparison

Option	Best for	Gap
OpenRouter (aggregator UI, not an ingestion API)	Browsing and routing through OpenRouter	Teams still need normalized export, filtering, and Apify scheduling
Internal FinOps spreadsheets	Manual reviews and finance notes	Slow updates, inconsistent capability fields, and weak automation
This actor (the normalized feed)	JSON snapshots, model admission checks, and recurring cost reviews	v1 has current snapshots only, not historical deltas

Use Cases

1. LLM gateway routing decisions by price, context, and capability.
2. AI agent platform model admission checks before enabling a new model.
3. FinOps cost per task budgeting for weekly model spend reviews.
4. AI engineering team weekly digest of price and capability changes.
5. Third party model marketplace listing data.

FAQ

How often should I run it? Weekly is enough for most teams. Daily is useful when you expose model choice to paying customers.

How accurate is capability detection? It follows OpenRouter fields. Vision uses architecture modality and input modalities. Tool and JSON support use advertised supported parameters.

Can provider data lag? Yes. OpenRouter is the primary normalized source, and direct provider pages are only fallback checks. Treat the feed as operational data, not a contract.

Does v1 include historical pricing? Not yet. /pricing-change returns a clear no history response. Schedule this actor and store snapshots if you need history today.

Can I get a refund for bad source data? Contact the actor owner with the run ID and affected model IDs. Refunds are reviewed case by case.

How do I request coverage changes? Send the model ID, provider URL, and desired field to the actor owner through Apify.