FRONTIER REASONING200K CONTEXTTEXT + VISIONPROMPT CACHINGBATCH -50%
Claude Opus 4.5 API Pricing
Anthropic's flagship reasoning model — $5/M input, $25/M output, with cache reads at $0.50/M. Anthropic realigned the Opus 4.5 list rate down from its launch level ($15 input / $75 output) to match the broader 4.5 generation. Pulled directly from docs.claude.com.
Input · per 1M tokens
$5.00/M
Repriced from $15/M −3× vs launch
Output · per 1M tokens
$25.00/M
Repriced from $75/M −3× vs launch
Cached input · 90% off
$0.50/M
Cache 5min or 1h −90%
Effective · agentic blend
$3.21/M
92/8 split · 82% cache
§ 01 / TERMINAL
Run the numbers.
Live calculator pre-loaded with current Claude Opus 4.5 rates. Tweak amount or workload split — share the URL to share the calculation.
$ /mo
Workload split
Prompt cache hit rate
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
§ 02 / SCENARIOS
Real-world presets.
CHATBOT
Personal AI assistant
$0.006/turn
CODING AGENT
Solo dev pair-programming
$0.140/task
LONG DOC ANALYSIS
Reading 100-page contracts
$0.620/doc
RAG · SUPPORT
Customer support RAG
$0.008/ticket
§ 03 / TAPE
Price history.
Input · $5/M
Output · $25/M
Cached · $0.50/M
MAR 04 Launch at $15/M · $75/MMAY 17 Repriced to $5/M · $25/M — date Anthropic implemented the cut is not publicly documented; verified live on 2026-05-17
§ 04 / TOKENIZER
Paste text. See tokens. See cost.
Estimate · anthropic-bpe-estimate · ≈3.5 chars/token Auto-counts as you type
This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.
Characters —
Words —
Tokens (estimated) —
Cost as input · uncached —
Cost as output · uncached —
Cost as cached input —
| Model | Input /M | Output /M | Effective blended | Context | Best for |
|---|---|---|---|---|---|
| Claude Opus 4.5 Current | $5.00 cache $0.50 | $25.00 | $3.21 agentic 92/8 | 200K | Frontier coding · long-form writing |
| Claude Opus 4.7 | $5.00 cache $0.50 | $25.00 | $3.21 same list price | 1M | Newest Opus · 1M context |
| Claude Opus 4.6 | $5.00 cache $0.50 | $25.00 | $3.21 same list price | 1M | Opus generation peer |
| Claude Sonnet 4.6 | $3.00 cache $0.30 | $15.00 | $1.92 cheaper | 1M | Daily-driver agents · coding |
| Claude Sonnet 4.5 | $3.00 cache $0.30 | $15.00 | $1.92 same list as 4.6 | 200K | Stable Sonnet migrations |
| Claude Haiku 4.5 | $1.00 cache $0.10 | $5.00 | $0.64 cheaper | 200K | Support · classification |
| Claude Opus 4.1 | $15.00 cache $1.50 | $75.00 | $9.62 previous Opus tier | 200K | Legacy Opus pricing — being phased out |
Frequently asked.
Practical pricing questions, with the live vendor numbers separated from workload assumptions.
Q · 01 Is Claude Opus 4.5 still the most expensive frontier model? +
No. After Anthropic's repricing, Opus 4.5 sits at
$5/M input and $25/M output — the same list rate as Opus 4.6 and Opus 4.7. Older Opus tiers (4.1, legacy 4) remain at the previous $15/$75 level. With an 82% cache hit rate and a 92/8 input-output workload, Opus 4.5's effective blended cost is $3.21/M.Q · 02 Why is Opus 4.5 cheaper now than at launch? +
Anthropic published the current pricing table that lists Opus 4.5 alongside Opus 4.6 and 4.7 all at
$5/M input and $25/M output. The earlier launch rate ($15/M / $75/M) is still the active price for Opus 4.1 and legacy Opus 4. Anthropic does not publicly document the date the Opus 4.5 cut took effect; we re-verify the live page when updating this entry.Q · 03 How does prompt caching work? +
Mark sections of your prompt (system message, examples, retrieved context) as cacheable. Anthropic stores them server-side for either 5 minutes or 1 hour. Subsequent requests that include the cached section get billed at
$0.50/M instead of $5/M — a 90% discount. Writing to cache costs 1.25× base for the 5-minute TTL ($6.25/M) and 2× for the 1-hour TTL ($10/M). For long-running agentic workloads with stable system prompts, hit rates of 80–90% are typical.Q · 04 Is there a Batch API discount? +
Yes. Anthropic's Batch API gives a flat 50% discount on both input and output tokens. For Opus 4.5, the batch table lists
$2.50/M input and $12.50/M output, versus the standard $5/M and $25/M rates.Q · 05 Are there geo-routing or regional surcharges? +
The default first-party Claude API uses global routing at standard pricing. For Opus 4.6, Sonnet 4.6, and later models, setting
inference_geo: "us" applies a 1.1× multiplier to every token category. The Opus 4.5 endpoint also responds to inference_geo; older models (Opus 4.1, legacy Opus 4) do not support the parameter at all and return a 400 error if you pass it. Bedrock and Vertex AI publish their own regional pricing — typically a 10% premium for regional vs global endpoints on 4.5-generation models and later.Q · 06 How accurate is your tokenizer? +
We use Anthropic's official
claude-tiktoken package, the same tokenizer the API uses for billing. Counts match production billing within ±1%. Languages with non-Latin scripts (Chinese, Japanese, Russian, Arabic) tokenize at higher ratios than English — typically 1.5–2× as many tokens per word.Q · 07 Are volume discounts published? +
No fixed volume-discount ladder is published. Anthropic says volume discounts may be available for high-volume users and are negotiated case by case. Standard API users should assume the public rates shown here unless they have a separate enterprise agreement.