PROVIDER GUIDE · VERIFIED 13 JUL 2026
Anthropic Claude API pricing, explained
Anthropic runs four price tiers across its Claude line, and one of them is currently on an introductory rate with a hard expiry date. Here's what each model actually costs, and what changes on 1 September 2026.
| Model | Input /1M | Output /1M | Context |
|---|---|---|---|
| Claude Haiku 4.5 | $1.00 | $5.00 | 200K |
| Claude Sonnet 5* | $2.00 | $10.00 | 1M |
| Claude Opus 4.8 | $5.00 | $25.00 | 1M |
| Claude Fable 5 | $10.00 | $50.00 | 1M |
* Introductory pricing through 31 Aug 2026. Standard rate from 1 Sep 2026: $3.00 / $15.00 per 1M.
The one date to put on your calendar
Claude Sonnet 5 is priced at $2.00/$10.00 per million tokens as an introductory rate. That changes to $3.00/$15.00 — a 50% increase on both input and output — starting 1 September 2026. If Sonnet 5 is your production model and your usage is meaningful, budget for the higher rate now rather than being surprised by a bill that jumps in September. For a workload running 500M input and 100M output tokens a month, that's the difference between a $1,000/month bill and a $1,500/month bill.
Picking the right tier for the job
- Haiku 4.5 — high-volume, low-complexity tasks: classification, extraction, routing, simple chat. Don't pay Sonnet or Opus rates for work Haiku handles fine.
- Sonnet 5 — the default production choice for most teams: strong reasoning at a mid-tier price, currently at introductory rates through August.
- Opus 4.8 — complex reasoning, long-horizon agentic tasks, work where being right matters more than being cheap.
- Fable 5 — Anthropic's most capable tier, priced accordingly. Reserve for tasks that genuinely need frontier-level capability.
The lever most people skip: prompt caching
If your prompts include a stable system prompt, a fixed set of instructions, or reference documents that repeat across calls, prompt caching can cut the cost of that repeated portion by roughly 90% on cache hits. For workloads with a large, static context reused across many calls — a coding assistant with a big codebase in context, a support bot with a fixed knowledge base — this is often a bigger saving than choosing a cheaper model tier.
Prices verified against platform.claude.com/docs/about-claude/pricing, 13 July 2026. See full methodology. Use the calculator to compare Claude against every other tracked provider for your own usage.