Claude Opus 4 Pricing
The most capable Claude model for complex reasoning, research, and agentic coding. Maximum intelligence, premium cost.
Input
$15.00
per MTok
Output
$75.00
per MTok
Cache Read
$1.50
per MTok (90% off)
Batch Input
$7.50
per MTok (50% off)
The Opus Premium: What 5x Buys You
Opus 4 costs exactly 5x more than Sonnet 4 for both input and output tokens. That is a significant premium, and it demands justification. So what are you paying for?
The Opus model is Anthropic's largest, most parameter-rich model. It excels in scenarios that require deep multi-step reasoning, handling ambiguity, synthesising information from large contexts, and producing nuanced, carefully structured output. In benchmarks and real-world testing, Opus consistently outperforms Sonnet on tasks that push the boundaries of language model capability.
Cost multiplier at a glance
vs Sonnet 4
5x
$15 vs $3 input
vs Haiku 3.5
18.75x
$15 vs $0.80 input
Output premium
5x
$75 vs $15 (Sonnet)
The key question is not “Is Opus better?” (it is) but “Is Opus 5x better for my specific task?” For most production workloads, the answer is no. Sonnet handles chatbots, content generation, standard code reviews, and data extraction with quality indistinguishable from Opus. The 5x premium pays for itself only on genuinely hard problems.
When Opus Is Worth the Premium
- +Legal analysis: contract review, compliance assessment, regulatory interpretation where errors have high consequences
- +Scientific research: literature synthesis, hypothesis evaluation, experimental design across complex domains
- +Complex multi-step reasoning: problems requiring 10+ logical steps, mathematical proofs, strategic planning
- +Code architecture: system design, large-scale refactoring decisions, security audit of complex codebases
- +Agentic workflows: autonomous coding sessions (Claude Code) where the model needs to plan, execute, and self-correct across many steps
- +High-stakes decision support: medical triage (non-diagnostic), financial analysis, risk assessment
When Opus Is NOT Worth It
- −Simple classification or routing: Haiku at $0.80/MTok delivers comparable accuracy for binary/multi-class tasks
- −Short content generation: blog posts, product descriptions, and emails rarely benefit from Opus over Sonnet
- −High-volume tasks: processing 100K+ requests/day at Opus prices is prohibitively expensive for most budgets
- −Chatbot conversations: Sonnet provides equally natural, helpful responses at 20% of the cost
- −Data extraction and formatting: structured output from documents does not require Opus-level reasoning
- −Prototype and development: use Sonnet during development, switch to Opus only if quality testing shows a need
Opus 4 Real-World Cost Examples
Five scenarios where Opus is the right model choice, with exact monthly costs.
Legal contract review
Full contract (~25K tokens) with detailed clause-by-clause analysis and risk flagging (~5K tokens output).
25,000 in + 5,000 out × 20 req/day × 30 days
Opus 4
$450.00
per month
Sonnet 4
$90.00
per month
Premium
+$360.00
per month
Scientific literature synthesis
Multiple paper abstracts and excerpts (~40K tokens) synthesised into a comprehensive research summary (~8K tokens).
40,000 in + 8,000 out × 10 req/day × 30 days
Opus 4
$360.00
per month
Sonnet 4
$72.00
per month
Premium
+$288.00
per month
Complex code architecture review
Multi-file codebase (~15K tokens) with architectural recommendations, refactoring plans, and security analysis (~6K tokens).
15,000 in + 6,000 out × 15 req/day × 30 days
Opus 4
$303.75
per month
Sonnet 4
$60.75
per month
Premium
+$243.00
per month
Agentic coding workflow (Claude Code)
Long-running agentic sessions (~50K tokens context including tool calls and results) with substantial code output (~10K tokens).
50,000 in + 10,000 out × 30 req/day × 30 days
Opus 4
$1,350.00
per month
Sonnet 4
$270.00
per month
Premium
+$1,080.00
per month
Strategic business analysis
Market data, financial figures, and competitive intel (~20K tokens) distilled into actionable strategy recommendations (~4K tokens).
20,000 in + 4,000 out × 5 req/day × 30 days
Opus 4
$90.00
per month
Sonnet 4
$18.00
per month
Premium
+$72.00
per month
Standard pricing without prompt caching or batch discounts. Enable prompt caching to reduce input costs by up to 90%.
Smart Model Routing: Use Opus Only When It Matters
The most cost-effective way to use Opus is not to use it for everything. Instead, implement a model routing strategy that sends each request to the cheapest model capable of handling it well.
Example routing strategy
Haiku 3.5
Classification, routing, entity extraction, formatting
Sonnet 4
Content generation, code review, Q&A, summarisation
Opus 4
Legal analysis, complex reasoning, architecture decisions
Estimated savings with routing
Assume 10,000 requests/day with an average 2,000 input + 800 output tokens each. Comparing all-Opus vs smart routing:
All Opus
$27,000.00
per month
Smart routing (60/30/10)
$5,184.00
per month
Savings: $21,816.00/month (81% reduction)
The simplest routing implementation uses Haiku as a classifier: send each incoming request to Haiku first, ask it to rate the complexity (low/medium/high), then route to the appropriate model. The Haiku classification call costs fractions of a cent and saves dollars on every request routed away from Opus. Read more in our cost reduction guide.
Opus vs Sonnet: Detailed Comparison
| Dimension | Opus 4 | Sonnet 4 |
|---|---|---|
| Input price | $15.00/MTok | $3.00/MTok |
| Output price | $75.00/MTok | $15.00/MTok |
| Cache read price | $1.50/MTok | $0.30/MTok |
| Batch input | $7.50/MTok | $1.50/MTok |
| Batch output | $37.50/MTok | $7.50/MTok |
| Context window | 200K tokens | 200K tokens |
| Speed | Slower, more thorough | Fast, production-ready |
| Reasoning depth | Deepest available | Strong, occasional gaps on hard problems |
| Coding | Best for architecture and complex refactoring | Excellent for everyday coding tasks |
| Recommended use | 5-10% of hardest tasks | 80-90% of production workload |