Claude Opus 5
Claude Opus 5 — Anthropic's most capable model for complex reasoning and long-document analysis.
| Type | Price / 1M tokens | Notes |
|---|---|---|
| Input | $5.00 | Per 1M input tokens |
| Output | $25.00 | Per 1M output tokens |
| Cache read | $0.50 | Cached prefix read-back |
| Cache write | $6.25 | Anthropic 5m TTL; 2× for 1h TTL |
At $15/M input and $75/M output, Opus 5 is priced for workloads where output quality directly drives revenue.
Best for
- Complex multi-step reasoning
- Legal and financial document analysis
- Advanced code generation and review
- Research synthesis across long documents
Not ideal for
- High-volume classification
- Cost-sensitive production inference
- Simple chatbots or FAQ systems
Real cost scenarios
| Scenario | Est. monthly cost | Breakdown |
|---|---|---|
| 1M tokens/day (moderate use) | $585 | 70% input ($315) + 30% output ($270) |
| RAG pipeline — 50k docs, 10k queries/day | $1,950 | Heavy context per query |
| Code review agent — 500 PRs/day | $3,375 | Long context, detailed output |
Estimates with typical token distributions. Use the Prompt Cost Calculator for your exact workload.
How to cut costs on Claude Opus 5
- Enable prompt caching on static system prompts and few-shot examples — saves up to 90% on prefix tokens.
- Route simple sub-tasks to Claude Haiku 4.5 — use Opus only for the steps that require it.
- Use batch API for non-real-time workloads — 50% cost reduction with same quality.
Calculate your Claude Opus 5 costs
Other Anthropic models
Compare Claude Opus 5
Frequently Asked Questions
When should I use Claude Opus 5 instead of Sonnet 5?
When task success rate directly affects revenue: complex contract review, multi-document synthesis, nuanced creative work where Sonnet 5 makes errors that Opus 5 doesn't. Run your eval suite on both before assuming you need Opus — Sonnet 5 handles 85%+ of professional tasks at 1/5 the cost.
Does Claude Opus 5 support prompt caching?
Yes. Anthropic's prompt caching supports Opus 5 with 5-minute and 1-hour TTL options. Write cost is 1.25× input price, read cost is 0.10× input price. Effective for system prompts >1,000 tokens that don't change between calls.
What is the context window for Claude Opus 5?
200K tokens (approximately 150,000 words or 600 pages). For most document analysis tasks, this is sufficient. For entire codebases or books, you'll need chunking or RAG.
Is Claude Opus 5 available for batch processing?
Yes — Anthropic's batch API supports Opus 5 at 50% off standard pricing ($7.50/M input, $37.50/M output). Results are delivered within 24 hours. Use batch for offline analytics, document processing, and eval runs.
Pricing verified 2026-08-19 from Anthropic official pricing. All prices in USD per 1M tokens. See methodology and price changelog.