Claude 5 Sonnet token & cost calculator

Claude 5 Sonnet is the current-generation workhorse — the model most production workloads should default to before reaching for Opus. It brings two changes that matter for budgeting: a 1,000,000-token context window (5× the 200K on Sonnet 4.5) and Anthropic's newer tokenizer, which counts about 30% more tokens for the same text. It's launching at an introductory $2 input / $10 output per million through August 31, 2026, then moves to standard $3/$15 on September 1.

The net effect today: Sonnet 5 is both newer and cheaper than Sonnet 4.5 — but the higher token count from the new tokenizer offsets part of the lower rate, so price a real prompt rather than assuming the intro discount flows straight through. After the intro window, the decision between 4.5 and 5 comes down to context window and quality, not price.

Expected response (output tokens)

Prompt

Client-side. Never uploaded.

0 / 1,000,000 charactersContext window: 1,000,000 tokens

Or start with an example

Total estimated cost

$0.010Claude 5 Sonnet

Tokens±2% approx

Input cost

$0.00

Output cost (est.)

$0.010

@ 1,024 response tokens

Context used

of 1,000,000

Verified 2026-07-06 · ±2%

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

Prompt uploads0Always 0 — by design

Outgoing requests0Analytics + page assets only — no prompt content

Cookies on this origin0Vercel Analytics + Clarity may set first-party cookies

localStorage keys0Theme preference + saved scenarios live here

Server endpoints1/api/og only — accepts title + subtitle, never prompt text

Inspect

Open DevTools → Network. Type into the calculator. No request bodies should contain your prompt text.

Pricing

Sonnet 5 is flat-priced — no tier surcharge across its million-token context. Today's rate is the introductory $2/$10 per million; on September 1, 2026 it becomes $3/$15. The dataAsOf date on this page tracks when we last reconciled against Anthropic's published rate.

Tier	Input $/M	Output $/M
All input	$2	$10
Context window	1,000,000 tokens

Verified against platform.claude.com on 2026-07-06.

Worked examples

Costs at the current introductory $2/$10 rate, from a short turn up to a large single-call input its million-token window now allows. Paste your own prompt above to price it exactly — and remember the September 1 move to $3/$15 raises each of these by half.

Scenario	Input	Output	Cost
Short chat turn A typical Q&A turn with a small system prompt.	800	400	<$0.01
RAG answer Retrieved context plus a grounded, structured response.	12,000	800	$0.032
Long-document Q&A A large input that wouldn't have fit in Sonnet 4.5's 200K window.	200,000	2,000	$0.420
Near-full context Close to the million-token ceiling in a single request.	900,000	4,000	$1.84

The input/output ratio is where Sonnet budgets are won or lost: at $2 input vs. $10 output (and $3 vs. $15 after September), a chat product whose typical turn is 800 input + 400 output spends most of its money on generated tokens. Clamp max_tokens, and cache stable prefixes — cached reads run at roughly a tenth of the input rate.

How is this counted?

Sonnet 5 uses Anthropic's newer tokenizer, which produces roughly 30% more tokens than the previous Claude encoding on the same text. Anthropic doesn't publish a client-side tokenizer, so we approximate: count with cl100k_base (gpt-tokenizer, MIT), then apply a ~1.3× calibration to track the newer tokenizer. That makes Sonnet 5 estimates rougher than for Sonnet 4.5 — budget ±10%, not ±2% — so treat the number as a planning figure and reconcile against the API response headers. Inputs over 50,000 characters run in a Web Worker.

FAQ

Why is Claude 5 Sonnet cheaper than Claude 4.5 Sonnet right now?

Introductory pricing. Sonnet 5 launched at a promotional $2 input / $10 output per million, in effect through August 31, 2026. On September 1, 2026 it moves to standard pricing of $3/$15 — the same rate as Sonnet 4.5. So today Sonnet 5 is both newer and cheaper; after the intro window it matches 4.5 on price while keeping the larger context window.

How is Claude 5 Sonnet different from Claude 4.5 Sonnet?

Two things that matter for cost planning: Sonnet 5 has a 1,000,000-token context window (5× the 200,000 on Sonnet 4.5), and it uses Anthropic's newer tokenizer, which packs about 30% more tokens into the same text. So a prompt that fit in 4.5 will report a higher token count — and cost — on Sonnet 5, even before the September price change.

How accurate is the token count?

Sonnet 5 uses Anthropic's newer tokenizer (~30% more tokens than the previous encoding on the same text). We approximate it by applying a ~1.3× calibration to the cl100k_base count. That's a solid planning estimate, but the multiplier varies with content, so treat the count as roughly ±10% rather than the ±2% we quote for older Claude models. Reconcile against the API response headers for billing.

What is the context window?

1,000,000 tokens — the full million-token window, matching Opus 4.8 and 5× what Sonnet 4.5 offered. The calculator warns you when input alone would exceed it.

When should I still reach for Opus?

Sonnet 5 is the default for production workloads. Move up to Opus 4.8 only for the slice of genuinely hard requests where a lift in reasoning quality pays for the roughly 2.5× price step; move down to Haiku 4.5 for cheap, high-volume classification. Route by difficulty.

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 approximate for Claude and Gemini — within ±2–3% for most, wider for the newer-tokenizer Claude models (Sonnet 5, Opus 4.7/4.8).

The natural comparison set: Claude 4.5 Sonnet (the prior-generation Sonnet, 200K context, standard $3/$15), Claude 4.8 Opus (the flagship, when the task needs stronger reasoning), and Claude 4.5 Haiku (the budget tier for high-volume, latency-sensitive work).