Anthropic Claude API Pricing (February 2026)
Current pricing for all Claude models. Anthropic offers significant discounts through prompt caching. Prices shown are per 1 million tokens.
| Model | Input Price | Cached Input | Output Price |
|---|---|---|---|
| Claude Opus 4.5Flagship | $12.00 / 1M | $1.20 / 1M (90% off) | $60.00 / 1M |
| Claude Sonnet 4.5Recommended | $3.00 / 1M | $0.30 / 1M (90% off) | $15.00 / 1M |
| Claude Haiku 4.5Best Value | $0.25 / 1M | $0.025 / 1M (90% off) | $1.25 / 1M |
Pro Tip: 90% Savings with Prompt Caching
Anthropic offers up to 90% discount on cached prompt tokens. Add cache_control to your API requests to enable caching. Perfect for applications with consistent system prompts.
Claude vs GPT: Which is Cheaper?
Here's a direct comparison of similar-tier models from Anthropic and OpenAI:
| Category | Claude Model | OpenAI Model | Winner |
|---|---|---|---|
| Flagship | Opus 4.5: $12/$60 | GPT-5: $15/$45 | Claude (input) / GPT (output) |
| Balanced | Sonnet 4.5: $3/$15 | GPT-4o: $2.50/$10 | GPT-4o (both) |
| Budget | Haiku 4.5: $0.25/$1.25 | GPT-4o-mini: $0.15/$0.60 | GPT-4o-mini (both) |
| With Caching | Opus: $1.20 (cached) | GPT-5: $7.50 (cached) | Claude (90% vs 50% discount) |
Summary: For raw pricing, OpenAI is slightly cheaper. But if you use prompt caching heavily, Claude's 90% discount beats OpenAI's 50% discount, potentially making Claude more cost-effective for your use case.
Choosing Between Opus, Sonnet, and Haiku
Claude Opus 4.5. When to Use
- Complex multi-step reasoning tasks
- Advanced coding and architecture decisions
- Creative writing requiring nuance
- Tasks requiring extended thinking
Claude Sonnet 4.5. When to Use
- Most production workloads
- Standard coding assistance
- Document analysis and summarization
- General-purpose chat applications
Claude Haiku 4.5. When to Use
- Simple classification tasks
- Data extraction and parsing
- High-volume, low-complexity operations
- Quick responses where speed matters
Frequently Asked Questions
Which is cheaper: Claude or GPT?
It depends on the tier and whether you use caching. For raw prices, GPT is slightly cheaper. But Claude's 90% prompt caching discount (vs OpenAI's 50%) can make Claude more cost-effective for applications with repeated prompts.
What is prompt caching and how does it work?
Prompt caching stores frequently used prompt prefixes so you pay less on subsequent requests. Anthropic offers up to 90% discount on cached tokens. Enable it by adding cache_control to your API requests. The cache lasts 5 minutes by default.
How many tokens can Claude process?
All Claude 4.5 models support 200K token context windows, approximately 150,000 words or 300+ pages. This allows processing entire codebases or long documents in a single request.
How do I reduce my Claude API costs?
1) Use Haiku for simple tasks. 2) Enable prompt caching for 90% savings. 3) Optimize system prompts. 4) Use batch processing when possible. 5) Consider Sonnet over Opus unless you need the extra capability.