Claude 4.7 Opus token & cost calculator
Claude 4.7 Opus is Anthropic's premium reasoning model — the one you reach for when the task is genuinely hard, when correctness is the product, and when a 5× per-million markup over Sonnet is justified by the lift in answer quality. The pricing geometry is unforgiving: at $15 input / $75 output per million tokens, a routine workload that runs 100% on Opus will burn cash at a rate that surprises most teams in their first month.
The discipline that makes Opus economic is routing. Send the easy 95% of requests to Sonnet or Haiku and reserve Opus for the 5% that need it. If you can articulate why a specific request type benefits from Opus's reasoning — and you can't get the same quality out of Sonnet with a more careful prompt — then you have a use case. Otherwise this is the model that takes a profitable feature and makes it unprofitable.
Saved scenariosnone yet
Saved on this browser only — never uploaded. Up to 10 scenarios.
Tip: save a scenario when you have a prompt + model + response length you might revisit. Useful for sizing features before committing to a vendor.
Verify privacysince this page loaded — updates live
Open DevTools → Network. Type into the calculator. No request bodies should contain your prompt text.
Pricing
Opus is flat-priced. The output multiplier is the same 5× as Sonnet and Haiku — the family is internally consistent on that ratio — but the absolute numbers shift the budgeting math.
| Tier | Input $/M | Output $/M |
|---|---|---|
| All input | $15 | $75 |
| Context window | 200,000 tokens | |
Verified against www.anthropic.com on 2026-05-09.
Worked examples
These scenarios show what Opus costs at three realistic prompt sizes. A "short chat turn" on Opus costs ~$0.03 — fine for a one-off, but at 100,000 daily turns that's $3,000/day. A long-document Q&A is over $1.50 per request before you sum the upstream pipeline.
| Scenario | Input | Output | Cost |
|---|---|---|---|
Short chat turn A typical Q&A turn with a small system prompt. | 800 | 400 | $0.042 |
System prompt + tool spec A larger context window with a tool schema, single response. | 5,000 | 500 | $0.112 |
Long document Q&A A long-form input (e.g. transcript) with a structured response. | 50,000 | 1,500 | $0.863 |
The right way to read these numbers is as a budget for selective use. If your assistant fields 1,000 requests a day and 50 of them get routed to Opus, the long-doc scenario above implies $75/day on the premium tier — defensible if those 50 requests are the high-stakes ones that drive customer outcomes. If 1,000/1,000 hit Opus, the math doesn't work.
How is this counted?
We approximate Opus's tokenizer with cl100k_base (gpt-tokenizer, MIT). For high-cost models the ~2% drift between the public encoding and the live Claude tokenizer matters — a 5% over-estimate on Opus is a real number of dollars in production. Use this calculator to size budgets and reconcile actuals from the API response headers. Inputs over 50,000 characters run in a Web Worker.
FAQ
- When the model's reasoning quality is the product, not the plumbing. Use Opus for tasks where a 5–10% lift in correctness is worth a 5× price uplift over Sonnet: hard agentic coding, deep multi-document synthesis, ambiguous reasoning chains, anything that fails consistently on Sonnet. If you can solve the task on Sonnet with a tighter prompt or a different decomposition, that is almost always cheaper than reaching for Opus.
- Three controls: clamp max_tokens aggressively, route to Opus only after a Sonnet first-pass tags a request as "hard," and cache anything you can. Most production systems should serve <5% of requests on Opus; if your share is higher, your routing layer is wrong, not your prompts.
- Yes — within ~2% on typical inputs. For Opus specifically, where individual requests are expensive, that drift can translate to dollars per request, so treat the calculator as a planning tool and reconcile actual spend against API response headers in production.
- 200,000 tokens, same as the rest of the Claude 4.x family. The calculator warns you when input alone would exceed this — Anthropic rejects out-of-window requests before the model runs.
- They optimize for different things. Opus is priced for selective, high-stakes reasoning at moderate context lengths. Gemini 2.5 Pro is priced for working with 200k–1M-token inputs; its tiered pricing makes large inputs cheaper than Opus, but on hard reasoning at typical context lengths Opus wins on quality. Pick by workload, not by sticker price.
When is Opus actually worth it?
How do I avoid runaway Opus bills?
Is the input token approximation reliable?
What is the context window?
How does it compare to Gemini 2.5 Pro's long-context tier?
Compare against every other model
To see this exact prompt scored against every supported model, sorted by total cost, paste it into the home calculator and toggle Compare across all models. Numbers are exact for OpenAI and within ±2–3% for Claude and Gemini.
Related models
The natural comparison set: Sonnet (the model most workloads should default to before reaching for Opus), Haiku (the budget option for the routing-layer below Opus), and Gemini 2.5 Pro (the cross-vendor premium tier with very different long-context economics).