Claude Pricing 2026: Every Model, Every Tier, Full Breakdown

Last updated: July 25, 2026

Claude pricing is more complex than a single number. Four current models, five subscription tiers, prompt caching, batch discounts, fast mode, introductory rates, and US-only inference all affect what you actually pay. This guide covers every layer — verified against Anthropic’s official pricing documentation on July 25, 2026.

Quick answer: Claude API pricing in July 2026 runs from $1/$5 per MTok (Haiku 4.5) to $10/$50 per MTok (Fable 5). Opus 5 at $5/$25 is the sweet spot for most production work. Sonnet 5 at $3/$15 (introductory $2/$10 through August 31) is the best value for high-volume tasks. Prompt caching cuts input costs by up to 90%, and the Batch API gives a flat 50% discount.

API pricing: full breakdown

All prices below are per million tokens (MTok). Sonnet 5 introductory pricing runs through August 31, 2026.

US-only inference is available at 1.1x standard pricing across all models. If your compliance requirements mandate US-based processing, add 10% to every figure above.

Legacy model pricing

Older models remain available but are not recommended for new work.

If you are still on Opus 4.8, the upgrade to Opus 5 is free in price and massive in capability. If you are on Sonnet 4.6, Sonnet 5’s introductory pricing makes the switch even easier.

Subscription pricing: Claude app plans

Claude’s subscription tiers determine which models you can use in the Claude app (Claude.ai), Claude Code, and Claude Cowork.

What each tier includes

Free gives you access to Claude Sonnet 5 with usage limits. It is the best way to evaluate Claude before paying. You cannot access Opus 5 or Fable 5 on the Free plan.

Pro ($20/month) raises usage limits significantly and adds access to Opus 5 — the strongest model available on Pro. Sonnet 5 remains the default. Pro is the right tier for individual professionals who use Claude daily for writing, research, and coding.

Max ($100/month or $200/month) makes Opus 5 the default model and provides substantially higher usage limits. The $200 tier offers the highest capacity. Fable 5 access is available through usage credits. Max is built for power users, heavy Claude Code workflows, and anyone who hits Pro limits regularly.

Team ($30/user/month) adds collaboration features, admin controls, and shared workspaces on top of Pro-level model access. It is designed for small teams that need centralized billing and policy controls.

Enterprise offers custom pricing, all available models, advanced security and compliance features, and dedicated support. Contact Anthropic sales for current Enterprise rates and terms.

Plan packaging, model availability, and usage credits can change. Check Anthropic’s pricing page for the latest tier details.

Prompt caching: how it works and what you save

Prompt caching is the single most effective way to reduce Claude API costs. If your prompts share a common prefix — system instructions, reference documents, codebase context — caching lets you pay the full input price only once and then reuse that cached content across subsequent requests.

How caching works

1. Cache write: The first request writes the shared prefix to the cache. You pay a premium over the base input price — 1.25x for a 5-minute TTL or 2x for a 1-hour TTL.
2. Cache hit: Subsequent requests that match the cached prefix pay only 10% of the base input price. This is where the savings come from.
3. TTL refresh: Each cache hit refreshes the time-to-live, so active conversations keep the cache warm without re-writing.

Savings by model

The math is simple: if your system prompt and reference context are 10,000 tokens and you make 100 requests, caching turns 1,000,000 input tokens into 10,000 write tokens plus 990,000 hit tokens. At Opus 5 pricing, that is roughly $5.50 instead of $50 — a 90% reduction on the input side.

When to use caching

• Long system prompts with detailed instructions or persona definitions
• Reference documents that stay constant across a session (legal text, specs, style guides)
• Codebase context in coding agents where the same files are referenced repeatedly
• Multi-turn conversations where the conversation history grows with each turn
• Batch processing where the same template is applied to many inputs

Caching is less useful for one-off queries with no shared prefix or for prompts where the entire input changes every time.

Batch API: 50% off for non-urgent work

The Batch API gives a flat 50% discount on both input and output tokens. The tradeoff is latency: batch requests are processed asynchronously, with results returned within 24 hours rather than in real time.

When to use the Batch API

• Large-scale content generation — product descriptions, summaries, translations
• Data extraction and classification — processing thousands of documents overnight
• Evaluation and testing — running benchmark suites without time pressure
• Backfill operations — generating embeddings, metadata, or annotations at scale

The Batch API is not suitable for interactive conversations, real-time agents, or any workflow where a user is waiting for a response.

Fast mode: 2x cost for ~2.5x speed

Fast mode is currently available on Opus 5 only. It runs the model at approximately 2.5x the default speed at twice the base price.

Fast mode is available on the Claude Platform and through usage credits in Claude Code. Use it for time-sensitive work where latency matters more than cost — live coding sessions, interactive debugging, or customer-facing applications where response time affects the experience.

For most batch and background work, standard speed is the better choice.

Cost optimization strategies

Paying the sticker price on every token is the most expensive way to use Claude. Here are the strategies that actually move the needle.

1. Model routing

The most effective strategy: start with the cheapest model that can handle the task, and escalate only on failure.

A practical setup: Sonnet 5 attempts the task. Tests or validators check the result. If the task fails or confidence is low, route to Opus 5. A human reviews high-risk outputs. This gives you better cost-performance than using Opus for everything.

2. Prompt caching

As covered above, caching cuts input costs by up to 90%. If your application has any repeated prefix — and most do — enable caching. The 5-minute TTL is sufficient for active sessions; the 1-hour TTL is better for longer workflows with gaps between requests.

3. Batch API for non-urgent work

If the user is not waiting, use the Batch API. A flat 50% discount with no quality tradeoff is the easiest optimization available.

4. Effort settings

Claude models support the effort parameter. Lower effort settings trade depth for speed and token consumption. A simple Q&A task does not need max effort. Reserve high and max effort for tasks where accuracy is critical and failures are expensive.

5. Right-size your context

The 1M-token context window is powerful, but stuffing it with irrelevant context wastes tokens and can degrade quality. Send only what the model needs. Use caching for the parts that repeat.

6. Use Sonnet 5 introductory pricing now

Sonnet 5’s introductory rate ($2/$10 through August 31, 2026) is the best value in the Claude lineup. If you have been waiting to migrate from Sonnet 4.6 or to test Claude for a new workload, this is the window.

Claude pricing vs OpenAI vs Google

How does Claude’s pricing compare to the other major providers?

*GPT-5.6 Sol applies a 2x input surcharge above 272K tokens. Claude’s 1M-token context has no such surcharge — a significant advantage for long-context workloads.

Key differences

• Context window pricing: Claude charges the same rate regardless of context length (up to 1M tokens). GPT-5.6 Sol doubles input pricing above 272K tokens. For long-document analysis, Claude is often cheaper in practice.
• Caching: All three providers offer prompt caching, but the discount structures differ. Claude’s 90% cache-hit discount is among the most aggressive.
• Batch discounts: Claude’s flat 50% batch discount is straightforward. Check current OpenAI and Google batch pricing for comparison.
• Introductory pricing: Sonnet 5’s $2/$10 intro rate (through August 31) is temporarily one of the most competitive prices in the premium-model segment.

For the detailed model-by-model comparison, see GPT-5.6 Sol vs Claude Fable 5.

Which Claude model should you use?

The best metric is cost per successful task, not price per token. Opus 5’s efficiency gains — fewer retries, fewer tokens, better self-verification — mean it can be cheaper in practice than a lower-priced model that needs more iterations.

FAQ