Anthropic Claude API Pricing Sheet & Token Rates
Anthropic Claude models lead the industry in autonomous software engineering, hybrid reasoning, and computer use. Explore token rates, prompt caching discounts (up to 90% off), and Batch API economics.
Anthropic Claude Monthly Token Spend Simulator
Calculate monthly bills across Claude 3.7 Sonnet and Claude 3.5 Haiku with prompt caching discounts.
Production Pricing & Model Tier Rates
Verified rates, context capacities, typical inference latency, and recommended enterprise workloads.
| Model / Tier | Input Rate (1M) | Output Rate (1M) | Context Window | Latency / Throughput | Recommended Workload |
|---|---|---|---|---|---|
| Claude 3.7 Sonnet | $3.000 ($0.300 cached) | $15.000 | 200,000 | Fast / Thinking | Premier agentic coding, hybrid thinking reasoning, computer use |
| Claude 3.5 Haiku | $0.800 ($0.080 cached) | $4.000 | 200,000 | Sub-200ms TTFT | Ultra-fast classification, lightweight coding, high-volume agents |
| Claude 3 Opus | $15.000 ($1.500 cached) | $75.000 | 200,000 | Legacy flagship | Deep philosophical and academic writing (largely superseded by 3.7) |
| Message Batches API | 50% Discount | 50% Discount | 200,000 | Within 24 hours | Asynchronous document processing, offline data evaluation |
âš¡ Architectural Deep Dive & Unit Economics Analysis
Anthropic has established itself as the technical leader in software engineering and agentic workflows. Their flagship Claude 3.7 Sonnet combines standard rapid inference with a dynamic extended thinking budget, allowing the model to reason through thousands of internal tokens to score a record-shattering 70.3% on SWE-bench Verified.
To optimize production economics, Anthropic offers Prompt Caching with a 5-minute time-to-live (TTL). While writing to the cache carries a 25% markup ($3.75/M), subsequent reads yield an astonishing 90% discount ($0.30 per million tokens). For coding agents repeatedly rescanning a 100k-token repository or customer service bots with heavy system instructions, prompt caching slashes token spend by up to 75%.
Production Workloads & Monthly Cost Modeling
Evaluating practical unit economics across real-world application architectures.
Autonomous Software Engineer (Cline / Cursor)
Pair programming with 40M context tokens (frequent cache hits) and 6M output tokens. Prompt caching reduces monthly cost from $210 to $105/dev.
Batch Code Refactoring (Message Batches API)
Migrating 2,000 Python files to modern typed async code within 24 hours. The 50% Message Batches discount cuts invoice from $800 to $400.
High-Volume Document Classification (Haiku)
Evaluating 50M tokens of incoming legal documents. Claude 3.5 Haiku processes this for $40 input + $20 output = $60.00 total.
import anthropic
client = anthropic.Anthropic()
def solve_with_claude_37(repo_context: str, task: str):
"""Leverage 90% prompt caching and dynamic thinking on Claude 3.7."""
response = client.messages.create(
model="claude-3-7-sonnet-20250219",
max_tokens=8192,
thinking={"type": "enabled", "budget_tokens": 4096},
messages=[
{
"role": "user",
"content": [
{
"type": "text",
"text": f"Repository Context:\n{repo_context}",
# Explicit cache breakpoint (90% discount on cache hits)
"cache_control": {"type": "ephemeral"}
},
{"type": "text", "text": task}
]
}
]
)
return response.contentOperational Gotchas & Failure Modes
Engineering realities, throttling thresholds, and hidden surcharge traps to monitor.
Related Models, Showdowns & Benchmarks
Cross-reference provider rates against hosted models, direct head-to-head battles, and calculators.
Claude 3.7 vs GPT-4o Showdown
Compare hybrid thinking reasoning against OpenAI flagship.
Claude 3.7 vs DeepSeek-R1
Compare frontier reasoning flags between Anthropic and DeepSeek.
DeepSeek-V3 vs Claude 3.5 Sonnet
Compare coding performance and token unit economics.
Prompt Caching ROI Calculator
Simulate exact cache hit break-evens and monthly savings.
Frequently Asked Questions
What is the token pricing for Claude 3.7 Sonnet?
Claude 3.7 Sonnet is priced at $3.00 per million input tokens ($0.30 on prompt cache hits) and $15.00 per million output tokens (including internal thinking tokens).
How does Anthropic prompt caching work?
By adding cache_control: {'type': 'ephemeral'} to your prompt messages, Anthropic caches the prompt prefix for 5 minutes. Initial cache write incurs a 25% surcharge ($3.75/M), but subsequent reads receive a 90% discount ($0.30/M).
What is Claude 3.7 Sonnet's thinking budget?
Thinking mode allows Claude 3.7 to generate extended reasoning tokens before producing an answer, scoring 70.3% on SWE-bench. You can set budget_tokens from 1,024 to 128,000 tokens.
What is the Message Batches API discount?
Anthropic offers a 50% discount across all input and output tokens for asynchronous batch jobs completed within 24 hours.
What is the pricing for Claude 3.5 Haiku?
Claude 3.5 Haiku is $0.80 per million input tokens ($0.08 on cache hits) and $4.00 per million output tokens, with a 200,000 token context window.
Does Anthropic train models on customer API data?
No. Anthropic does not train foundation models on customer API inputs or outputs, and offers enterprise HIPAA Business Associate Agreements (BAAs).
Claude 3.7 & 3.5 Pricing Architecture: Prompt Caching & Thinking Token Budgets
Anthropic's flagship Claude models—Claude 3.7 Sonnet, Claude 3.5 Sonnet, and Claude 3.5 Haiku—lead the industry in coding, instruction adherence, and complex tool-calling agents. Operating at scale on Anthropic requires mastering their Prompt Caching pricing mechanics: writing cache checkpoints incurs a 25% surcharge ($3.75/M tokens), but subsequent read hits receive a massive 90% discount ($0.30 per million tokens). For Claude 3.7 Sonnet, developers can enable extended thinking, which dynamically generates internal reasoning tokens billed at the standard output rate of $15.00/M tokens.
Anthropic Production Rate Card (Per Million Tokens)
| Model | Standard Base Input | Cache Write (Initial) | Cache Read (90% Off) | Output / Thinking Tokens |
|---|---|---|---|---|
| Claude 3.7 Sonnet (Hybrid) | $3.00 / M | $3.75 / M | $0.30 / M | $15.00 / M |
| Claude 3.5 Sonnet | $3.00 / M | $3.75 / M | $0.30 / M | $15.00 / M |
| Claude 3.5 Haiku | $0.80 / M | $1.00 / M | $0.08 / M | $4.00 / M |
Claude 3.7 Extended Thinking Budget Controller
import Anthropic from '@anthropic-ai/sdk';
const anthropic = new Anthropic();
// Clamp thinking token budget to exactly 2,048 tokens to prevent bill runaway
export async function runReasoningTask(systemInstructions: string, userTask: string) {
return await anthropic.messages.create({
model: 'claude-3-7-sonnet-20250219',
max_tokens: 4096,
thinking: {
type: 'enabled',
budget_tokens: 2048 // Strict clamp limits reasoning token cost to $0.03 max
},
system: [{ type: 'text', text: systemInstructions, cache_control: { type: 'ephemeral' } }],
messages: [{ role: 'user', content: userTask }]
});
}Engineering Q&A: Claude API Optimization
How long do Anthropic prompt cache checkpoints stay active?
Anthropic caches operate on a 5-minute sliding Time-to-Live (TTL). Every cache read request refreshes the 5-minute timer. For continuous background agent workers, a keepalive ping every 4.5 minutes maintains active cached state indefinitely.
What is the minimum prompt length to enable caching on Claude?
Prompts must contain at least 1,024 tokens for Claude 3.5 Sonnet and Claude 3.7 Sonnet, and 2,048 tokens for Claude 3 Opus. Smaller prompts will not be cached and execute at standard base rates.