Claude Opus 4.8 pricing
Anthropic ·
claude-opus-4-8 ·
1M context
· 59th cheapest of 64 current models
What Claude Opus 4.8 costs
| Rate | Per 1M tokens | Per 1K tokens | What it covers |
|---|---|---|---|
| Input | $5 | $0.005 | Every token you send: prompt, history, documents. |
| Output | $25 | $0.025 | Every token the model generates, including hidden reasoning tokens. |
| Cached input | $0.5 | $0.0005 | Input served from the provider’s prompt cache. |
| Cache write | $6.25 | $0.00625 | Charged once, to put a prefix into the cache. |
What that means in practice
The same three workloads are costed on every model page, so these numbers are directly comparable across the catalog. No prompt caching and no batch discount is assumed — the figures are what you pay before you optimise anything.
Support chatbot
2,500 six-turn conversations a day
$20,138 / month
$0.269 per conversation · $245,006 a year
An 800-token system prompt, 400-token questions and 900-token answers, over six turns — each turn re-sending everything before it.
Document summariser
1,000 single-shot summaries a day
$2,430 / month
$0.081 per conversation · $29,565 a year
One 12,000-token document in, an 800-token summary out, no conversation history. The case where input dominates the bill.
Coding agent
200 twelve-turn sessions a day
$14,040 / month
$2.34 per conversation · $170,820 a year
A 3,000-token system prompt, 1,500-token instructions and 2,500-token responses over twelve turns. Long sessions are where the quadratic cost of re-sent history stops being theoretical.
Change the assumptions in the calculator
Cheaper models in the catalog
Nothing cheaper is scored as highly as Claude Opus 4.8, so these are simply the cheaper options, best-scored first. Trading down here is a real trade.
| Model | Input | Output | Blended saving | |
|---|---|---|---|---|
| Claude Sonnet 5 | $3 | $15 | 40% | |
| Gemini 3.1 Pro | $2 | $12 | 55% | side by side |
| GPT-5.4 | $2.5 | $15 | 44% | |
| GPT-5.6 Terra | $2 | $12 | 55% | |
| Claude Sonnet 4.6 | $3 | $15 | 40% |
Head to head
Specification
| Context window | 1M tokens |
|---|---|
| Maximum output | 128K tokens |
| Provider | Anthropic |
| Status | Current |
| Reasoning model | Yes |
| Vision | Yes |
| Token counting | Estimated at ~3.6 characters per token (Estimate — Anthropic's tokenizer is server-side; ratio calibrated on English prose) |
| Batch API | 50% off both rates for non-interactive work |
Where these numbers come from
Last checked against its source on 2026-08-01, and the published rates last actually moved on 2026-08-02. Source of record: the vendor’s own published pricing.
The vendor page a human checked
Regional and data-residency premiums, priority tiers, server-side tool fees and negotiated discounts are not included. Prices change without notice; this page is rebuilt every time the catalog does.