How accurate are these LLM API prices?
Prices are comparison data transformed from public pricing metadata. Verify provider pages before production routing or purchasing decisions.
Cost intelligence for AI infrastructure
Find the cheapest LLM API route.
Compare prices, cache discounts, prompt costs, and model routes. Before paying for cloud tokens, check whether the model fits on your own GPU. Local AI companion tool Can my GPU run this LLM? Check VRAM fit, model size, quantization, and local-vs-cloud fallback. Open local GPU checker Read update
Quick route
Find the right model route for the job.
First choose the use case. Then choose whether you want the best model, the cheapest route, best value, or your own model.
1. Use case
2. Optimize for
The result uses the live model catalog, selected token settings, and rule-based quality/use-case scoring.
Cheapest
Open GPU check
Loading route...
The router is loading model pricing data.
- One run
- $0.000000
- 10/day
- $0.000000
- 300/month
- $0.000000
- Context
- n/a
Why this helps
The route shows the cheapest fitting model and scales it to a workflow budget.
Top picks
Loading picks...
Cost estimate
Current prompt cost
$0.000000
- Words
- 0
- Tokens
- 0
- Cached
- 0
Prompt cost engine Edit prompt and cost settings
Token Calculator
Prompt cost engine
Token approximation: 1 word equals roughly 1.3 tokens.
Best Route
Recommended API route
One clear route first. Open the settings when the job needs different requirements.
Loading recommendation...
Route settings Adjust job and compare alternatives
Loading routes...
Live Route API
Test the endpoint
Live Route API
Test the endpoint
Endpoint
POST /api/recommend-route
Cache share
0%
Local / private route: sensitive and private jobs force local-first and apply conservative cloud exclusions. Check LocalAI or an approved private environment before sending raw data to a cloud model.
Server recommendation waits for a request.
Arbitrage Matrix Full leaderboard Live
Model views
Loading models...
| Model | Provider | Context | Input |
|---|---|---|---|
| Loading models... | |||
Data source
Freshness, source and currency view
Models tracked
Loading
Checked
unknown
Prices are comparison data. Verify provider pages before production routing or purchasing decisions.
Planning score is a rule-based heuristic derived from price tier, context window, capabilities, and model name signals. It is not an external benchmark, Elo rating, or LMArena score. Local open-weight models can serve as a practical cost fallback when frontier model pricing rises or availability changes; use this as a planning signal, not a performance guarantee.
Token Waste Check
Find overpaid agent routes
Compare the selected model with cheaper SLM, budget, and local-open routes before a multi-step agent spends tokens.
Cache share
50%
Fine-tune tokens and constraints
Loading agent workload estimate...
Cheaper matching routes
| Route | Cost | Context |
|---|---|---|
| Loading routes... | ||
Model Intelligence
Loading model data...
Selected model metadata updates with the calculator.
Model details Prices, context and capabilities
Output / 1M
$0.00
Cache Read / 1M
$0.00
Context Window
0
Max Output
0
Features
source_key: loading
checked: loading
Context Calculator
Document fit and cost matrix
Paste text or choose a preset to compare document fit, output limits, and one-time versus cached analysis costs.
Fits
0
Output capped
0
Too large
0
Context matrix Compare all models
| Model | Fit | Context Used | One-time | Cached repeat |
|---|---|---|---|---|
| Enter text to compare all models. | ||||
Token approximation and cache costs
The calculator uses a simple token approximation: 1 word ≈ 1.3 tokens. It is not an exact tokenizer simulation, but it is useful for fast cost comparisons across LLM APIs and for estimating prompt length before production use.
Cache read costs describe lower prices for prompt segments that have already been cached. With providers such as OpenAI or Anthropic, reused system prompts, long contexts, or repeated prefixes can cost significantly less than entirely new input tokens.
FAQ
LLM pricing questions
What is cached input pricing?
It is a lower price for repeated prompt segments, reused system prompts, document prefixes, or stable context blocks when a provider supports prompt caching.
Which model is cheapest for long context?
It depends on prompt size, output tokens, cache share, and required capabilities. Use the context matrix and Best Route cards for the current job.
Is apiroute.dev affiliated with model providers?
No. Provider and model names are used referentially for API pricing, capability, and compatibility comparisons.
For AI Agents
Machine-readable pricing and routing layer
Use these endpoints directly in agents, crawlers, and workflow tools. No HTML table scraping required.
Best default JSON endpoint for current model pricing, capabilities, freshness, and source metadata.
Model list only, useful when an agent already knows the metadata policy.
OpenAPI 3.1 spec for tool registration in agents and workflow systems.
Compact crawler guide for discovering the pricing API and agent-readable docs.
Route recommendation guide
Static machine guide for the Best Route V1 scoring inputs and route labels.
Token Waste Check contract
Agent-readable guide for detecting overpaid repeated prompts and cheaper SLM/budget routes.
{
"tool": "apiroute_prices",
"openapi": "https://apiroute.dev/openapi.yaml",
"default_endpoint": "https://apiroute.dev/api/live-prices",
"routing_guide": "https://apiroute.dev/api/route-recommendation-guide",
"recommend_endpoint": "https://apiroute.dev/api/recommend-route",
"token_waste_contract": "https://apiroute.dev/api/token-waste-check"
}Business Layer
Commercial options without paid rankings
apiroute.dev can test monetization through alerts, provider sponsorships, and premium data access while keeping the comparison table independent.
Price change alerts
A simple waitlist for teams that want alerts when model prices, context windows, or cache discounts change.
Join alert waitlistAI/ML API provider option
One OpenAI-compatible API for many hosted models. Use it as a provider option when you want to test model routing without integrating every vendor separately.
Affiliate link. This does not affect model rankings, calculator results, or route recommendations.
Try AI/ML APIPremium data access
Future paid access could add higher refresh frequency, pricing history, diff alerts, and machine-readable change logs.
Request API detailsClearly labeled sponsor slots
Provider placements can be sold only if clearly labeled. Core price rankings stay sorted by data, not payment.
Discuss sponsorshipCommercial metadata is available for agents and partners as JSON. It describes what is testable now and what is intentionally not sold.
Market Radar