Written by Espen, who helps business owners install AI systems using open-source tools. What is a Chief AI Officer? →

Anthropic Claude API Pricing in February 2026 — Every Model, Every Tier

The complete, no-fluff breakdown of what Claude actually costs when you call it through the API. Opus 4.6, Sonnet 4.5, Haiku 4.5 — with real numbers, real-world cost examples, and comparisons to OpenAI and Google.

Claude API pricing in February 2026: Opus 4.6 costs $5/$25 per million tokens (input/output), Sonnet 4.5 costs $3/$15, and Haiku 4.5 costs $1/$5. Prompt caching cuts input costs by 90%. Batch processing saves 50%. Those are the numbers — the rest of this guide explains what they mean for your wallet.

Big update: Anthropic dropped Opus pricing by 67% with the 4.6 release. The old Opus 4 cost $15/$75 — the new Opus 4.6 costs just $5/$25. If you're still on Opus 4, switching to 4.6 saves you massive money and gets you a better model.

This guide covers the Anthropic API specifically — the developer-facing, pay-per-token service you access through api.anthropic.com. If you're looking for Claude Code (the CLI tool) pricing, see our Claude Code pricing guide. For OpenClaw costs, check the OpenClaw pricing guide.

Claude API Pricing 2026: All Models at a Glance

Anthropic offers three model tiers through the API. Each targets a different use case and budget. Note: prompts over 200K tokens cost more on Opus and Sonnet.

Long-Prompt Pricing (Over 200K Tokens)

If your prompts exceed 200K tokens, Opus 4.6 and Sonnet 4.5 charge more:

Prompt Caching: The Biggest Cost Saver

Prompt caching is the single most impactful way to reduce your Claude API bill. If you're sending the same system prompt, few-shot examples, or documents repeatedly, you're leaving money on the table.

Here's how it works: the first time you send a prompt with caching enabled, you pay a 25% surcharge to write the content into the cache. Every subsequent request that reuses that cached content pays only 10% of the normal input price. The cache has a 5-minute TTL (time-to-live) that refreshes with each hit.

Prompt Caching: Before vs. After

Let's say you run a chatbot with a 4,000-token system prompt, handling 100 requests/hour on Sonnet 4.5:

When caching saves you money

• Chatbot with a long system prompt — Cache the system prompt. Every user message only pays the cache-read rate on that portion.
• RAG with fixed context — Cache your retrieved documents when the same docs serve multiple queries.
• Few-shot classification — Cache your examples. Only the new input is charged at full price.

Batch Processing: 50% Off Everything

If your workload doesn't need real-time responses, the Message Batches API gives you a flat 50% discount on both input and output tokens. Batches are processed within 24 hours.

Batch processing is ideal for data labeling, content generation pipelines, bulk summarization, and any workflow where you can queue requests and process results later.

Extended Thinking Costs

Claude's extended thinking feature (where the model "thinks" step-by-step before responding) uses additional tokens. These thinking tokens are charged at the output token rate — which is the expensive side.

For Sonnet 4.5, that means thinking tokens cost $15/MTok. For Opus 4.6, $25/MTok. This adds up fast on complex reasoning tasks. A request that generates 5,000 thinking tokens on Opus 4.6 costs an extra $0.125 just for the reasoning.

Claude vs OpenAI API Cost: February 2026 Comparison

Here's how Claude stacks up against the competition in February 2026. We're comparing models at roughly equivalent capability tiers.

Flagship / Premium Tier

The pricing landscape shifted dramatically with Opus 4.6. At $5/$25, it's now 76% cheaper than GPT-5.2 Pro on input and 85% cheaper on output. Gemini 3 Pro remains the cheapest flagship option, but the gap has narrowed considerably.

Workhorse / Mid Tier

This is the tier most developers care about. GPT-5.2 edges out Sonnet 4.5 on input pricing ($1.75 vs $3.00) and is roughly comparable on output ($14 vs $15). Gemini 3 Flash undercuts both significantly — but each model has different strengths in coding, reasoning, and instruction-following.

Budget / Speed Tier

At the budget tier, GPT-5 mini is the cheapest option. Haiku 4.5 is the most expensive of the three — but it consistently delivers higher quality output. For high-volume classification, extraction, or routing tasks, test all three on your specific workload before choosing based on price alone.

Real-World Cost Examples

Abstract token prices are hard to reason about. Here are five specific production scenarios with exact math, all using Sonnet 4.5:

Cost Optimization Cheatsheet

Claude API vs Claude Pro vs Claude Max: Which Should You Use?

Anthropic offers three ways to use Claude. Here's when each makes sense:

When Claude Pro ($20/mo) beats the API

If you're an individual using Claude for writing, research, brainstorming, and analysis through the chat interface, Pro is almost certainly cheaper. Rough rule of thumb: if you'd use fewer than ~700,000 output tokens per month on Sonnet 4.5 via the API ($10.50), Pro's unlimited-ish access at $20/month is the better deal because you also get Opus access, projects, memory, and Claude Code.

When Claude Max ($100–$200/mo) makes sense

Max is for power users who hit Pro's usage limits regularly — especially heavy Claude Code users. The $100/month tier gives you 5x Pro usage, and $200/month gives you 20x. If you're a developer using Claude Code for hours daily, Max pays for itself vs. API costs for equivalent token volume.

When the API wins

The API is the only choice when you need: programmatic access, custom system prompts, tool use, automated pipelines, batch processing, or integration into your own products. There's no substitute for the API when you're building software.

Claude API Free Tier

Anthropic does not offer a free API tier with ongoing free tokens. However:

• Free claude.ai usage — You can use Claude for free through the web interface with limited daily messages. No API access.
• Pay-as-you-go API — The API charges only for tokens consumed. No minimum spend. You can start with just a few dollars of credits.
• Usage tier system — New accounts start at tier 1 with lower rate limits. Limits increase as you spend more ($0 → $4,000+ monthly spend tiers).

If you want to test Claude without spending money, use the free web tier first. When you're ready to build, the API has no setup fees — you pay only for what you use.

Legacy Model Pricing

If you're still using older Claude models, here's what they cost. Consider upgrading — newer models are often cheaper and better:

OpenClaw + Claude API: Efficient AI Automation

OpenClaw is a personal AI automation platform that connects to the Claude API (among other providers) to run your AI workflows 24/7. Here's how it keeps costs efficient:

• Smart model routing — OpenClaw can use cheaper models (Haiku) for simple tasks and upgrade to Sonnet or Opus only when complexity requires it.
• Prompt caching by default — System prompts and agent context are cached automatically, saving up to 90% on repeated tokens.
• Batching where possible — Background tasks like email triage and content generation use the batch API for 50% savings.
• Token-aware scheduling — OpenClaw tracks token usage and can throttle or schedule tasks to stay within budget.
• Multi-provider support — Route tasks to the cheapest capable model across Anthropic, OpenAI, and Google.

A typical OpenClaw setup running email monitoring, calendar management, and daily briefings on Sonnet 4.5 costs roughly $15–40/month in API fees — less than a Claude Pro subscription, but with full automation capabilities.

How to Access the Claude API

Getting started takes about 2 minutes:

1. Create an account at console.anthropic.com
2. Add a payment method — credit card or prepaid credits
3. Generate an API key in the dashboard
4. Set spending limits — Anthropic lets you set hard monthly caps
5. Make your first call:

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-sonnet-4-5-20241219",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Hello, Claude"}]
  }'

Anthropic uses a prepaid/postpaid billing model. You can buy credits upfront or get billed monthly. There are usage-based rate limits that increase as you spend more (tiers range from $0 to $4,000+ in monthly spend).

What's Changed Recently

Anthropic has been actively adjusting pricing and releasing new models. Key recent changes:

• Claude Opus 4.6 launched — The flagship model now costs $5/$25, a 67% price drop from Opus 4's $15/$75
• Claude Sonnet 4.5 launched — Maintains $3/$15 pricing with improved performance
• Claude Haiku 4.5 launched — Upgraded from Haiku 3.5 ($0.80/$4) to $1/$5, slightly more expensive but significantly more capable
• Tiered pricing for long prompts — Prompts over 200K tokens now cost 2x on input (1.5x on output) for Opus and Sonnet
• Extended thinking generally available across all models
• Web search tool — $10 per 1,000 searches, new capability
• Code execution — 50 free hours/day per org, then $0.05/hour

Bottom Line

For most developers building with the Claude API in February 2026, the play is clear: Sonnet 4.5 with prompt caching. It gives you excellent quality at an effective input rate that's cheaper than Haiku 4.5's standard pricing. Use Haiku 4.5 for high-volume, simple tasks where you don't need caching. And the new Opus 4.6 at $5/$25 is finally affordable enough for production — consider it for agent workflows and complex coding tasks.

Compared to OpenAI and Google, Claude's pricing is competitive. Opus 4.6's price cut makes it dramatically cheaper than GPT-5.2 Pro. At the mid tier, Sonnet 4.5 is slightly pricier than GPT-5.2 on input but comparable on output. Google's Gemini remains the budget king. The real differentiator isn't price — it's which model performs best on your specific tasks.