/ writing / automation

OpenRouter Models Tracker

2026-08-23 · cynix · 3 min read

Model pricing and availability on OpenRouter shift weekly. The catalog is a clean, keyless public API — 422 models right now, each with per-token pricing, context length, architecture, modality, reasoning support, and knowledge cutoff. Nobody on Apify packages that as a trackable, diffable feed. This actor does: it snapshots the whole catalog into RAG-ready records and tells you what changed since the last run.

We scoped this as the Tier-2 follow-up to the MCP Registry Indexer. The original Tier-2 idea was an LM Arena leaderboard tracker, but LMArena sits behind Cloudflare (403 on the data API, JS-rendered HTML) — building on that means scraping a bot-wall for unparsed HTML, which fails the publish-readiness gate. OpenRouter's API is the durable, keyless alternative in the same "AI model intelligence" space.

What it does

Indexes every model: model_id, name, context_length, prompt_price_per_mtok, completion_price_per_mtok, architecture, modality, reasoning, knowledge_cutoff, provider, capabilities.
Filters by capability (reasoning, vision, function_calling, json_mode, web_search, audio, video), minimum context, or max price.
Detects change — stores a snapshot in actor KV and emits change records for NEW / REMOVED / PRICE DROP / PRICE RISE models versus the previous run. First run establishes the baseline and emits nothing.

Input

{
  "onlyCapabilities": ["reasoning"],
  "minContextLength": 200000,
  "maxPricePerMTok": 2,
  "changeMonitor": true
}

Output (a filtered model record)

{
  "kind": "model",
  "model_id": "deepseek/deepseek-v4-flash-vision",
  "name": "DeepSeek: DeepSeek V4 Flash Vision",
  "context_length": 1048576,
  "prompt_price_per_mtok": 0.22,
  "reasoning": true,
  "capabilities": ["reasoning", "function_calling", "json_mode"],
  "change": ""
}

A real run with the filter above returned 195 reasoning models at ≥200k context under $2/MTok. A second run (catalog unchanged) returned 422 models and zero change records — the delta logic doesn't emit false positives.

Why publishable, why first-mover

Reads only publicly published model metadata from OpenRouter's open API — no personal data, no login, no ToS conflict. The Store has ~250 "openrouter" results, but none extract the catalog as a clean, filterable, change-tracked feed. Scheduling it daily gives you a model-market watch without touching the API yourself.

OpenRouter Models Tracker on Apify. Pay-per-event: a small per-run start fee plus a per-record item fee. Keyless source, no proxy cost.

← back to writing