@reaatech/agent-budget-pricing

Status: Pre-1.0 — APIs may change in minor versions. Pin to a specific version in production.

LLM pricing table management with LRU-cached lookups, model name normalization, and cost computation. Ships with built-in pricing for Anthropic, OpenAI, Google, and AWS Bedrock.

Installation

npm install @reaatech/agent-budget-pricing
# or
pnpm add @reaatech/agent-budget-pricing

Feature Overview

• 4 providers built in — Anthropic (Claude), OpenAI (GPT/o-series), Google (Gemini), AWS Bedrock (multi-model)
• Token cost computation — computeCost(inputTokens, outputTokens, modelId) returns exact dollar cost, with optional cache read/write token accounting
• Cost estimation — estimateCost(modelId, estimatedInputTokens) for pre-flight budget checks
• Model name normalization — ModelNormalizer maps provider aliases to canonical model IDs
• LRU cache with TTL — configurable cache TTL (default 1 hour) avoids repeated table lookups
• Programmatic table loading — loadTable(provider, models) for custom or updated pricing
• File-based overrides — load pricing from JSON files to override or extend built-in tables
• Dual ESM/CJS output — works with import and require

Quick Start

import { PricingEngine } from '@reaatech/agent-budget-pricing';

const pricing = new PricingEngine({ cacheTtlMs: 3600_000 });

const cost = pricing.computeCost(500, 200, 'claude-sonnet-4', 'anthropic');
// → 0.0045  (500/1M × $3 + 200/1M × $15)

const estimate = pricing.estimateCost('claude-sonnet-4', 1000, 'anthropic');
// → 0.003   (1000/1M × $3 input, assuming 0.2 output ratio)

API Reference

PricingEngine

| Method | Description | | ---------------------------------------------------------------------- | ------------------------------------------------------------ | | lookup(modelId, provider?) | Look up PriceEntry for a model. Returns null if unknown. | | computeCost(inputTokens, outputTokens, modelId, provider?, options?) | Calculate dollar cost for actual token usage | | estimateCost(modelId, estimatedInputTokens, provider?, outputRatio?) | Estimate cost before the call (default output ratio: 0.2) | | loadTable(provider, models) | Load or replace pricing data for a provider | | clearCache() | Invalidate the LRU cache |

Constructor

new PricingEngine(options?: {
  cacheTtlMs?: number;    // default: 3600_000 (1 hour)
  normalize?: boolean;     // default: true — enable model name normalization
})

ModelNormalizer

import { ModelNormalizer } from '@reaatech/agent-budget-pricing';

const normalizer = new ModelNormalizer();
normalizer.registerAlias('anthropic', 'claude-3-opus', 'claude-opus-4-1');
normalizer.normalize('claude-3-opus', 'anthropic'); // → "claude-opus-4-1"

Built-in Pricing Tables

Anthropic

| Model | Input (per 1M tokens) | Output (per 1M tokens) | | ---------------- | --------------------- | ---------------------- | | Claude Opus 4-1 | $15.00 | $75.00 | | Claude Sonnet 4 | $3.00 | $15.00 | | Claude Haiku 3.5 | $0.25 | $1.25 |

OpenAI

| Model | Input (per 1M tokens) | Output (per 1M tokens) | | ----------- | --------------------- | ---------------------- | | GPT-4 | $30.00 | $60.00 | | GPT-4 Turbo | $10.00 | $30.00 | | GPT-4 Mini | $0.15 | $0.60 | | o1 | $15.00 | $60.00 | | o3 | $15.00 | $60.00 |

Google Gemini

| Model | Input (per 1M tokens) | Output (per 1M tokens) | | ---------------- | --------------------- | ---------------------- | | Gemini 2.0 Flash | $0.10 | $0.40 | | Gemini 2.0 Pro | $1.25 | $5.00 |

AWS Bedrock

| Model | Input (per 1M tokens) | Output (per 1M tokens) | | --------------- | --------------------- | ---------------------- | | Claude 3 Sonnet | $3.00 | $15.00 | | Claude 3 Haiku | $0.25 | $1.25 | | Llama 3 70B | $0.50 | $0.70 | | Mistral Large | $4.00 | $12.00 |

Usage Patterns

Custom Pricing Tables

import { PricingEngine } from '@reaatech/agent-budget-pricing';

const pricing = new PricingEngine();
pricing.loadTable('my-provider', {
  'my-model-v1': {
    inputPricePerMillion: 1.5,
    outputPricePerMillion: 5.0,
    cacheReadPricePerMillion: 0.15,
    cacheWritePricePerMillion: 1.875,
  },
  'my-model-v2': { inputPricePerMillion: 2.0, outputPricePerMillion: 8.0 },
});

const cost = pricing.computeCost(1000, 500, 'my-model-v1', 'my-provider', {
  cacheReadTokens: 250,
  cacheWriteTokens: 100,
});

Integration with Budget Check

import { PricingEngine } from '@reaatech/agent-budget-pricing';
import { BudgetController } from '@reaatech/agent-budget-engine';

const pricing = new PricingEngine();
const controller = new BudgetController({ spendTracker, pricing });

// BudgetController uses pricing.estimateCost() for pre-flight checks
const result = await controller.check({
  scopeType: BudgetScope.User,
  scopeKey: 'user-42',
  estimatedCost: pricing.estimateCost('claude-opus-4-1', 2000, 'anthropic'),
  modelId: 'claude-opus-4-1',
});

Related Packages

• @reaatech/agent-budget-types — Core types (BudgetScope, SpendEntry)
• @reaatech/agent-budget-engine — Enforcement engine
• @reaatech/agent-budget-spend-tracker — Spend tracking

License

MIT