TL;DR — Anthropic's current Claude model pricing in 2026: Opus 4.8 at $5/$25, Sonnet 4.6 at $3/$15, and Haiku 4.5 at $1/$5 per million input/output tokens. Opus 4.5+ received a massive 67% price cut from the $15/$75 of Opus 4.0/4.1. Opus and Sonnet now include 1M-token context at no surcharge. Even with the price drop, 60-70% of requests routed to Opus don't require its reasoning depth. Teams using ClawRouters to auto-route each request to the right Claude tier are cutting their Anthropic spend by 40-60% with no quality loss.
Anthropic's Claude family spans three tiers — Opus, Sonnet, and Haiku — each sharing the same API format, the same discount programs, and the same 5:1 output-to-input price ratio. The only differences are capability, speed, and cost. This guide breaks down every current Claude model's pricing with real production numbers, compares them to competitors, and shows exactly how to optimize your spend.
Anthropic Claude Model Pricing Table (May 2026)
All prices are per million tokens (MTok). These are the official rates from Anthropic's pricing page.
| Model | Input (/1M) | Output (/1M) | Context | Max Output | Cache Read | Batch API | Best For | |-------|------------|-------------|---------|-----------|-----------|-----------|----------| | Claude Opus 4.8 | $5.00 | $25.00 | 1M | 128K | $0.50/M | $2.50/$12.50 | Most complex reasoning, agentic coding | | Claude Opus 4.7 | $5.00 | $25.00 | 1M | 128K | $0.50/M | $2.50/$12.50 | Complex reasoning (previous gen) | | Claude Opus 4.6 | $5.00 | $25.00 | 1M | 128K | $0.50/M | $2.50/$12.50 | Complex reasoning (previous gen) | | Claude Sonnet 4.6 | $3.00 | $15.00 | 1M | 64K | $0.30/M | $1.50/$7.50 | Coding, writing, general analysis | | Claude Haiku 4.5 | $1.00 | $5.00 | 200K | 64K | $0.10/M | $0.50/$2.50 | Classification, extraction, simple Q&A |
Legacy models (deprecated or retired):
| Model | Input (/1M) | Output (/1M) | Status | |-------|------------|-------------|--------| | Claude Opus 4.1 | $15.00 | $75.00 | Available (legacy pricing) | | Claude Opus 4.0 | $15.00 | $75.00 | Deprecated, retiring June 15, 2026 | | Claude Sonnet 4.0 | $3.00 | $15.00 | Deprecated, retiring June 15, 2026 | | Claude Haiku 3.5 | $0.80 | $4.00 | Retired (Bedrock/Vertex only) |
Key Pricing Rules
Every current Claude model follows these pricing patterns:
- 5:1 output-to-input ratio — Output tokens always cost exactly 5x input tokens. This ratio has been consistent across every Claude generation.
- Prompt caching — 5-minute cache writes cost 1.25x base input. 1-hour cache writes cost 2x. Cache reads cost just 0.1x (90% savings). A cache hit pays for itself after a single read.
- Batch API — 50% off both input and output tokens for async workloads processed within 24 hours.
- Fast mode — Opus 4.6/4.7 at $30/$150. Opus 4.8 at $10/$50. Faster output at premium pricing.
- No free tier — Anthropic provides starter credits for new accounts but no ongoing free usage.
The Opus Price Drop: What Changed
The biggest pricing shift of 2026 was Opus 4.5 dropping from $15/$75 to $5/$25 — a 67% reduction. Anthropic maintained this lower price through Opus 4.6, 4.7, and the latest 4.8 release. This means Opus is now just 1.67x the cost of Sonnet on a per-token basis, down from 5x under the old pricing.
If you're still budgeting based on the $15/$75 Opus 4.0 rates, your cost projections are 3x too high. Legacy Opus 4.1 still uses the old pricing — upgrading to Opus 4.5+ saves 67% with better performance.
Real-World Monthly Costs by Claude Model
Raw per-token pricing only tells part of the story. Here's what each model costs at typical production volumes:
| Daily Volume | Opus 4.8 /month | Sonnet 4.6 /month | Haiku 4.5 /month | |-------------|-----------------|-------------------|------------------| | 500K in + 500K out | $450 | $270 | $90 | | 2M in + 2M out | $1,800 | $1,080 | $360 | | 5M in + 5M out | $4,500 | $2,700 | $900 | | 5M in + 5M out (Batch) | $2,250 | $1,350 | $450 | | 5M in + 5M out (Cached + Batch) | ~$1,350 | ~$810 | ~$270 |
Cost Per Request
For a typical API request (800 input tokens, 400 output tokens):
| Model | Input Cost | Output Cost | Total/Request | 10K Requests/day | |-------|-----------|-------------|---------------|-------------------| | Opus 4.8 | $0.004 | $0.010 | $0.014 | $4,200/mo | | Sonnet 4.6 | $0.0024 | $0.006 | $0.0084 | $2,520/mo | | Haiku 4.5 | $0.0008 | $0.002 | $0.0028 | $840/mo |
Even after the Opus price drop, sending a classification task to Opus costs 5x more than Haiku — with no difference in output quality for that task type.
When to Use Each Claude Model
Claude Opus 4.8 — The Reasoning Flagship ($5/$25)
Opus 4.8 is Anthropic's most capable model. With 1M-token context and 128K max output, it handles:
- Multi-step mathematical proofs and formal verification
- Complex agentic coding with autonomous tool use
- Legal document analysis requiring nuanced interpretation
- Research synthesis across dozens of sources
- High-autonomy tasks requiring 10+ reasoning steps
Cost reality: Opus is worth every token for genuinely complex work. But ClawRouters routing data shows only 30-40% of requests sent to Opus actually need its reasoning depth.
Claude Sonnet 4.6 — The Production Workhorse ($3/$15)
Sonnet 4.6 is where most production traffic should go. At 40% cheaper than Opus, it excels at:
- Code generation, debugging, and review
- Content writing and editing
- Data analysis and summarization
- Multi-turn conversation
- Moderate complexity reasoning (5-8 steps)
Value insight: Sonnet 4.6 now includes 1M context at no surcharge — the same context window as Opus.
Claude Haiku 4.5 — Speed and Efficiency ($1/$5)
Haiku 4.5 is dramatically underestimated. At 5x cheaper than Opus and 3x cheaper than Sonnet, it handles:
- Intent classification and entity extraction
- Simple Q&A and FAQ responses
- Content moderation and filtering
- Data formatting and transformation
- Routing decisions (selecting which model should handle a request)
Speed advantage: Haiku's sub-200ms first-token latency makes it ideal for latency-sensitive pipelines. Many teams use Haiku as the classifier layer in a routing system — it decides which requests need Sonnet or Opus, saving the cost of sending everything to expensive models.
Claude Pricing vs. Competitors (May 2026)
| Provider / Model | Input (/1M) | Output (/1M) | vs. Opus | vs. Sonnet | Notes | |-----------------|-------------|--------------|----------|------------|-------| | Claude Opus 4.8 | $5.00 | $25.00 | — | — | 1M context, 128K output | | Claude Sonnet 4.6 | $3.00 | $15.00 | 40% cheaper | — | 1M context | | Claude Haiku 4.5 | $1.00 | $5.00 | 5x cheaper | 3x cheaper | 200K context | | OpenAI GPT-5.5 | $5.00 | $30.00 | Same input, 20% more output | — | | | OpenAI GPT-5.4 | $2.50 | $15.00 | 50% cheaper | Same output | | | Google Gemini 3 Pro | $1.25 | $5.00 | 5x cheaper | 3x cheaper | | | Google Gemini 3 Flash | $0.075 | $0.30 | 83x cheaper | 50x cheaper | | | DeepSeek V4 Flash | $0.14 | $0.28 | 89x cheaper | 54x cheaper | |
What the Numbers Reveal
After the Opus price cut, Claude is much more competitive. Opus 4.8 now matches GPT-5.5 on input price and costs less on output ($25 vs $30). Sonnet remains a strong mid-tier option. But budget options like Gemini Flash and DeepSeek V4 Flash are still 50-80x cheaper.
The question isn't whether Claude is expensive — it's whether you're sending the right requests to the right model tier. A classification request doesn't need Opus, and a multi-step research task shouldn't go to Haiku.
How to Optimize Your Claude API Costs
Strategy 1: Smart Model Routing (40-60% Savings)
The single most effective optimization is to stop using one model for everything. Production traffic analysis shows:
- 30-40% of requests need Opus-level reasoning
- 40-50% of requests are well-served by Sonnet
- 15-25% of requests can be handled by Haiku with identical quality
ClawRouters automates this decision. Set model="auto" and the router analyzes each prompt in real-time to select the optimal model — including cheaper non-Claude alternatives when appropriate. No code changes beyond swapping the base URL.
Example: A team spending $4,500/month on all-Opus traffic switched to ClawRouters auto-routing. Result: 20% stayed on Opus, 50% routed to Sonnet, 30% went to Haiku or cheaper alternatives. Monthly bill dropped to $1,900 — a 58% reduction.
Strategy 2: Prompt Caching (Up to 90% Input Savings)
If you send repeated system prompts or context blocks, prompt caching is essential:
- 5-minute cache write: 1.25x base input price
- 1-hour cache write: 2x base input price
- Cache read: 0.1x base input price (90% savings)
For Opus at $5/M input, cached reads drop to $0.50/M — the same as Haiku's cache read price.
Strategy 3: Batch API (50% Off)
For workloads without real-time requirements — data processing, content generation, batch analysis — the Batch API halves your costs:
| Model | Standard Output | Batch Output | Savings at 5M out/day | |-------|----------------|-------------|----------------------| | Opus 4.8 | $25.00/M | $12.50/M | $1,875/mo | | Sonnet 4.6 | $15.00/M | $7.50/M | $1,125/mo | | Haiku 4.5 | $5.00/M | $2.50/M | $375/mo |
Strategy 4: Stack All Three
Maximum savings come from combining these strategies:
- Route each request to the cheapest capable model (ClawRouters handles this automatically)
- Cache repeated system prompts and context across all models
- Batch non-real-time workloads
Combined, these can reduce your total Claude bill by 60-80% compared to sending everything to a single model without optimization.
Claude Model Pricing History and Outlook
How Claude Pricing Has Evolved
| Date | Event | Pricing Impact | |------|-------|---------------| | Mid 2025 | Claude Opus 4 + Sonnet 4 launch | $15/$75 Opus, $3/$15 Sonnet established | | Late 2025 | Claude Opus 4.5 launch | Opus drops to $5/$25 (67% cut) | | Early 2026 | Opus 4.6 + Sonnet 4.6 launch | Same pricing, 1M context added for both | | April 2026 | Opus 4.7 launch | $5/$25 maintained | | May 2026 | Opus 4.8 (latest) | $5/$25 maintained, improved capability |
The pattern: Anthropic now delivers more capability at stable prices. The Opus 4.5 price reset was a one-time correction; since then, pricing has held steady through three model generations.
Will Anthropic Cut Prices Again?
The $5/$25 Opus pricing appears stable through at least Q3 2026. With GPT-5.5 priced at $5/$30 and no frontier competitor significantly undercutting, Anthropic has little pressure to cut further. The smart move: optimize with routing and caching now. For a deeper analysis, see our how to reduce LLM API costs guide.
Frequently Asked Questions
What is Anthropic Claude model pricing in 2026?
Anthropic offers three active Claude model tiers in 2026: Claude Opus 4.8 at $5 input / $25 output per million tokens, Claude Sonnet 4.6 at $3 input / $15 output, and Claude Haiku 4.5 at $1 input / $5 output. All models qualify for 50% Batch API discounts and prompt caching (cache reads at 10% of base input price). Opus and Sonnet 4.6 include 1M-token context windows.
How much did Claude Opus pricing change in 2026?
Claude Opus received a major price cut with the 4.5 generation: from $15/$75 per million tokens (Opus 4.0/4.1) down to $5/$25 — a 67% reduction. This lower pricing has been maintained through Opus 4.6, 4.7, and the latest 4.8 release.
Which Claude model offers the best value?
Claude Sonnet 4.6 at $3/$15 per million tokens is the best value for most production workloads. It handles coding, writing, and analysis at 40% less than Opus. For maximum savings, use an LLM router like ClawRouters to automatically select the cheapest model for each request.
How much does Claude Opus 4.8 cost per month?
At typical production volume (2M input + 2M output tokens per day), Claude Opus 4.8 costs approximately $1,800/month. Using prompt caching and Batch API can reduce this to under $1,000/month. Smart routing through ClawRouters can cut it further to $700-$1,100/month by redirecting simpler requests to Sonnet or Haiku.
Is Claude more expensive than GPT or Gemini?
It depends on the tier. Claude Opus 4.8 ($5/$25) now matches GPT-5.5 ($5/$30) on input and is cheaper on output. Sonnet 4.6 ($3/$15) matches GPT-5.4 on output but costs more on input. Haiku 4.5 ($1/$5) is significantly pricier than budget options like Gemini 3 Flash ($0.075/$0.30) or DeepSeek V4 Flash ($0.14/$0.28).
How can I reduce my Anthropic Claude API costs?
Three strategies: (1) Use smart model routing via ClawRouters to send each request to the cheapest capable Claude model — saves 40-60%. (2) Enable prompt caching to cut repeated input costs by up to 90%. (3) Use the Batch API for async workloads to get 50% off. Combined, these reduce your Claude bill by 60-80%.
What is the cheapest way to use Claude in 2026?
Combine Claude Haiku 4.5 ($1/$5) with the Batch API ($0.50/$2.50) and prompt caching (cache reads at $0.10/M input). For workloads needing higher capability on specific requests, use ClawRouters to route only complex tasks to Sonnet or Opus while keeping everything else on Haiku.
Does Claude have a 1M context window?
Yes. Claude Opus 4.6, 4.7, 4.8 and Sonnet 4.6 all support 1M-token context windows at standard per-token pricing — no surcharge for longer contexts. Haiku 4.5 has a 200K-token context window.