Live LLM pricing and specs API for routing agents.
Your model router needs to know what each LLM costs and supports right now, not what your training data remembers. This endpoint returns the current pricing list - USD per 1M input and output tokens, context window, vision and tool support, status - for the closed flagships, plus the trending open models, in one keyless call.
The call
One keyless POST returns the closed-model pricing rows plus the trending open models. search_ai_models matches model names (not prices); ai_models returns the open-model side from Hugging Face. Each row carries price-per-1M tokens, context window, vision and tool flags, status, an observed-at timestamp and an Ed25519 signature.
curl -s https://dynamicfeed.ai/v1/batch \
-H 'content-type: application/json' \
-d '{"calls":[
{"tool":"search_ai_models","args":{"query":"claude"}},
{"tool":"ai_models","args":{}}
]}'Sample response
{ "tool": "search_ai_models", "args": { "query": "haiku" } }
// or filter the registry: provider, max_input_price, min_context, needs_vision
// the same tools are exposed over the keyless MCP at https://dynamicfeed.ai/mcpWhy live data
A routing agent that hard-codes prices drifts the moment a provider ships a new tier or model. This feed pulls from a model-pricing registry refreshed on a cycle, so the prices and context windows you read are the ones in effect, each stamped with the source and the time it was last refreshed. Note what the signature does and does not mean: the Ed25519 signature is proof-of-existence and tamper-evidence - it proves Dynamic Feed reported this row at this time and that the bytes have not changed since. It is not a claim that the price is correct, current at your read, or safe to route on. Treat it as verify-not-trust: you can confirm the data is exactly what we served and check the observed-at age yourself, then make the routing call. Use search_ai_models for the closed flagships (GPT, Claude, Gemini, Grok) with pricing; use ai_models for the open ecosystem (Llama, Qwen, DeepSeek, Mistral, Gemma). search_ai_models matches model names, not prices, so query "haiku" or "gpt", then filter by max_input_price or min_context.
Use it for
- Pick the cheapest model that fits a request's context and vision needs at call time
- Keep a cost dashboard or budget guardrail current without scraping provider pages
- Detect when a new model or pricing tier ships and update routing rules automatically
- Ground an agent answer about model costs in a signed, timestamped row instead of stale memory
FAQ
Do I need an API key?
No. POST to https://dynamicfeed.ai/v1/batch or point an MCP client at https://dynamicfeed.ai/mcp - both are keyless and free for fair use.
What fields come back per model?
Provider, model id, input and output USD per 1M tokens, cached-input price, context window, max output tokens, modalities, vision and tool support, and status. search_ai_models also takes provider, max_input_price, min_context and needs_vision filters.
Does the signature mean the price is correct?
No. The Ed25519 signature is proof-of-existence and tamper-evidence - it proves we served these exact bytes at the stamped time. It is not a truth, accuracy or safety claim. Check the observed-at age and verify against the provider before routing real spend.