or-pricing

A CLI tool that shows realistic expected pricing for OpenRouter models by querying per-provider endpoints and weighting by uptime.

No API key required · Works via npx · Node 18+

Contents

• The Problem
• Install
• Quick Start
• Usage
  • price
  • search
  • compare
  • leaderboard
• Flags
• How Expected Pricing Works
• Limitations
• JSON Output
• APIs Used
• Contributing
• License

The Problem

OpenRouter's headline pricing shows the cheapest available provider, but the price you actually pay depends on which provider handles your request. Here's Kimi K2.5:

Headline:  $0.45 / $0.44 per M tokens  ← cheapest provider
Expected:  $0.56 / $2.67 per M tokens  ← weighted across all healthy providers

Some providers may be down or using different quantization levels, so the effective price often differs from the headline. or-pricing queries every provider's pricing, uptime, and status, then computes a weighted expected price.

Install

# Use directly via npx (zero install)
npx or-pricing search deepseek --json

# Or install globally
npm install -g or-pricing

# Or clone and build
git clone https://github.com/quinnblay/openrouter-tools.git
cd openrouter-tools
npm install && npm run build
node dist/index.js --help

No API key required — uses OpenRouter's public API. Requires Node 18+.

Quick Start

npx or-pricing price anthropic/claude-sonnet-4
npx or-pricing compare openai/gpt-4o anthropic/claude-sonnet-4 deepseek/deepseek-r1
npx or-pricing search deepseek
npx or-pricing leaderboard --refresh

Usage

or-pricing price [models...]        Per-provider pricing + weighted expected price  (alias: p)
or-pricing search <query>           Search models by name                          (alias: s)
or-pricing compare [models...]      Side-by-side pricing comparison                (alias: cmp)
or-pricing leaderboard              Top models by usage on OpenRouter              (alias: lb)

Both price and compare accept multiple models as arguments or via stdin (whitespace-separated):

or-pricing price anthropic/claude-sonnet-4 openai/gpt-4o                  # multiple args
echo "anthropic/claude-sonnet-4 openai/gpt-4o" | or-pricing compare       # stdin (space-separated)
echo -e "anthropic/claude-sonnet-4\nopenai/gpt-4o" | or-pricing compare   # stdin (newline-separated)

When pricing multiple models, endpoint queries are paced at ~0.3s apart to avoid rate limiting.

price — Detailed model pricing

or-pricing price kimi-k2.5

MoonshotAI: Kimi K2.5  (moonshotai/kimi-k2.5)
Context: 262,144 tokens

Headline:  $0.45 / $0.44 per M tokens (prompt/completion)
Expected:  $0.56 / $2.67 per M tokens (weighted by uptime, healthy only)

Provider Breakdown:
  Provider           Quant    Prompt/M   Compl/M    Uptime   Status
  Chutes             int4     $0.45      $0.44      97.0%    ok
  DeepInfra          unknown  $0.45      $2.25      98.4%    ok
  Inceptron          int4     $0.50      $2.40      85.5%    DOWN
  AtlasCloud         int4     $0.50      $2.60      98.8%    ok
  Together           unknown  $0.50      $2.80      75.9%    DOWN
  SiliconFlow        fp8      $0.55      $3.00      98.9%    ok
  Novita             unknown  $0.57      $2.85      98.3%    ok
  Parasail           int4     $0.60      $2.80      95.5%    ok
  Moonshot AI        int4     $0.60      $3.00      99.9%    ok
  Fireworks          unknown  $0.60      $3.00      98.2%    ok
  BaseTen            fp4      $0.60      $3.00      99.4%    ok
  Venice             unknown  $0.75      $3.75      99.3%    ok

Accepts full model IDs (moonshotai/kimi-k2.5) or fuzzy matches (kimi-k2.5, k2.5). Multiple matches prompt you to be more specific.

search — Find models

or-pricing search deepseek

16 model(s) matching 'deepseek':

  Model ID                                       Prompt/M    Compl/M     Context
  deepseek/deepseek-chat                         $0.3        $1.2        160K
  deepseek/deepseek-r1                           $0.7        $2.5        62K
  deepseek/deepseek-v3.2                         $0.25       $0.38       160K
  ...

compare — Side-by-side comparison

or-pricing compare anthropic/claude-sonnet-4 openai/gpt-4o deepseek/deepseek-v3.2

Model Comparison

  Model                                     Head P/M      Head C/M      Exp P/M       Exp C/M       Providers   Context
  Anthropic: Claude Sonnet 4                $3            $15           $3.00         $15.00        5/5         976K
  OpenAI: GPT-4o                            $2.5          $10           $2.50         $10.00        2/2         125K
  DeepSeek: DeepSeek V3.2                   $0.25         $0.38         $0.32         $0.63         7/9         160K

Shows headline vs. expected pricing, provider counts, and context length for each model.

leaderboard — Top models by usage

or-pricing leaderboard                                     # global OpenRouter leaderboard (from cache)
or-pricing leaderboard --refresh                           # scrape fresh global data
or-pricing leaderboard --app https://openclaw.ai/          # per-app leaderboard
or-pricing leaderboard --app https://openclaw.ai/ --refresh  # scrape fresh app data

Shows the top 20 models by token usage.

Without --app, shows the global OpenRouter leaderboard from openrouter.ai/rankings. With --app, shows per-app usage from the OpenRouter apps page. Each app gets its own cache file.

Data is fetched from OpenRouter's rankings pages (parsed from server-rendered HTML) and cached locally. You can also pipe leaderboard data from external tools:

echo '{"entries":[{"rank":1,"model":"Kimi K2.5","author":"moonshotai","tokens":"1.38T tokens"}]}' \
  | or-pricing leaderboard --update-cache

Flags

Global (all commands):

Flag	Description
--json	Output structured JSON instead of formatted tables
--no-color	Disable colored output (also respects $NO_COLOR)
--help, -h	Show help

Leaderboard only:

Flag	Description
--app <url>	Per-app leaderboard (e.g. --app https://openclaw.ai/)
--refresh	Fetch fresh leaderboard data instead of using cache
--update-cache	Accept leaderboard JSON from stdin to update cache

How Expected Pricing Works

For each provider serving a model, the OpenRouter endpoints API reports pricing, uptime (last 30 minutes), and status. The expected price algorithm:

1. Exclude unhealthy providers — status < 0 (degraded or DOWN) are shown in the table but excluded from the average
2. Weight by uptime — each healthy provider's weight is uptime / 100
3. Apply discounts — effective price = price * (1 - discount)
4. Weighted average — expected = sum(effective_price * weight) / sum(weight)

Provider status mapping:

• >= 0 → ok (included in expected price)
• -1 → degraded (excluded)
• <= -2 → DOWN (excluded)

Limitations

• Model list is cached for 5 minutes — not real-time pricing data
• Expected price is a weighted estimate, not a guarantee of what you'll be charged
• Designed for research and cost comparison, not billing integration

JSON Output

All commands support --json for programmatic use.

JSON output examples (click to expand)

price --json

{
  "id": "moonshotai/kimi-k2.5",
  "name": "MoonshotAI: Kimi K2.5",
  "context_length": 262144,
  "headline": {
    "prompt_per_m": 0.45,
    "completion_per_m": 0.44
  },
  "expected": {
    "prompt_per_m": 0.56,
    "completion_per_m": 2.67
  },
  "providers": [
    {
      "provider": "Chutes",
      "quantization": "int4",
      "prompt_per_m": 0.45,
      "completion_per_m": 0.44,
      "discount": 0,
      "status": 0,
      "uptime": 97.0
    }
  ]
}

search --json

[
  {
    "id": "moonshotai/kimi-k2.5",
    "name": "MoonshotAI: Kimi K2.5",
    "context": 262144,
    "prompt_per_m": 0.45,
    "completion_per_m": 0.44
  }
]

compare --json

[
  {
    "id": "moonshotai/kimi-k2.5",
    "name": "MoonshotAI: Kimi K2.5",
    "context": 262144,
    "headline_prompt": 0.45,
    "headline_completion": 0.44,
    "expected_prompt": 0.56,
    "expected_completion": 2.67,
    "providers": 12,
    "healthy": 10
  }
]

leaderboard --json

{
  "entries": [
    {
      "rank": 1,
      "model": "Kimi K2.5",
      "author": "moonshotai",
      "tokens": "1.38T tokens"
    }
  ],
  "cached_at": "2026-02-15T20:25:32Z"
}

APIs Used

Endpoint	Auth	Purpose
GET /api/v1/models	None	Model catalog, headline pricing
GET /api/v1/models/{author}/{slug}/endpoints	None	Per-provider pricing, uptime, status
openrouter.ai/rankings/trending	None (HTML fetch)	Leaderboard data

Model list responses are cached for 5 minutes in .cache/models.json.

Contributing

Contributions welcome. Before submitting a PR:

1. Run npm run typecheck and npm run build
2. Manually test affected commands (no test suite)
3. File an issue before starting major work

License

MIT