6 min read · updated 2026-10-31 · reviewed

Claude Opus API price in 2026: every Opus model, official vs prepaid

Claude Opus is the model people reach for when a task is hard, and its price is the first thing they check. Here are the current Opus prices at Anthropic and on Model.sale, worked examples for a single request and an agent step, and an honest note on the workloads where the official API costs less.

Editorial note

Written by Model.sale Engineering and reviewed against first-party documentation on 2026-10-31. Comparisons use the billing rules visible on the linked sources and do not claim universal model equivalence.

Official Anthropic prices

Claude Opus 5.5 (claude-opus-5-5): $4 per million input tokens, $20 per million output tokens, cache reads $0.20 and a 5-minute cache write $5 per million. Claude Opus 4.8: $5 input and $25 output. For comparison, Sonnet 5.5 is $2 / $10 and Fable 5.1, the tier above Opus, is $10 / $50.

These rates were checked on 2026-10-01 against Anthropic's pricing page; check it again before you budget, because Anthropic changes prices with new releases. Opus 5.5 is both newer and cheaper than Opus 4.8, so there is rarely a reason to stay on the older model.

Model.sale prices

Model.sale charges one rate per million charged tokens, with cached input counted at 40%. On 2026-10-31: claude-opus-5-5 $1.40; claude-opus-5, claude-opus-4-8, claude-opus-4-7 and claude-opus-4-6 $2.00; claude-fable-5-1 $3.40; claude-sonnet-5-5 $0.90. The balance is prepaid in USD from $5 with crypto or $10 by card, and it does not expire.

What one request costs

A chat-style request with 10,000 input and 1,000 output tokens: officially 0.01 × 4 + 0.001 × 20 = $0.06 on Opus 5.5; on Model.sale 11,000 charged tokens × $1.40 per million = $0.015, about 74% less.

A long agent step with 200,000 input tokens, 180,000 of them cached, and 2,000 output tokens: officially 0.02 × 4 + 0.18 × 0.20 + 0.002 × 20 = $0.156 plus any cache writes; on Model.sale 20,000 + 2,000 + 0.4 × 180,000 = 94,000 charged tokens = $0.132, only about 15% less: on cache-heavy steps the two are close.

When the official API is cheaper

Anthropic's cached input is very cheap ($0.20 per million), while Model.sale counts cached tokens at 40% of its flat rate ($0.56 per million on Opus 5.5). A workload that is almost entirely cache reads with very little output can therefore cost the same or less at Anthropic. The saving on Model.sale comes from output and fresh input, which Anthropic prices at $20 and $4.

A typical Claude Code session sits in between: on a medium refactor we measured about 36% saving on Opus 5.5 before cache-write charges. Run one real task and read the exact charge in the dashboard before committing a budget.

How to pay and start

Create an account, add funds (USDT and other crypto from $5, card, PayPal, Apple Pay or Google Pay from $10), create a key and point the client at Model.sale. For Claude Code set ANTHROPIC_BASE_URL=https://api.model.sale and ANTHROPIC_AUTH_TOKEN to the key; OpenAI-compatible tools use https://api.model.sale/v1.

Give the key a daily spend limit before an agent runs unattended. Opus burns budget fastest in long loops, and a prepaid wallet stops at zero instead of sending an invoice.

Sources and further reading

Checked on 2026-10-31. These links support the factual comparisons in this article; pricing and terms may change.

Use the numbers on your own workload

Open the cost calculator, choose a live model and enter a representative input/output mix before deciding.

Open cost calculator
← All guides