Skip to content
induwara.lk
Premium
induwara.lkAI · Developer tools

AI Reasoning Token Cost Calculator

Find the real cost of a reasoning-model API call. This tool adds the hidden reasoning (“thinking”) tokens that o-series, Claude, and Gemini bill at the output rate, so you see the true per-call and monthly bill, and how much more it is than a naive token estimate. No signup, sources cited below.

By Induwara AshinsanaUpdated Jun 6, 2026
Reasoning token cost
Anthropic-verified rate

Your prompt, system message, tools, and context.

The answer you actually see in the response.

Estimate from "Medium" effort, edit to your real value.

How many of these calls you make in a month.

Example workloads
True cost / call
$0.135
incl. 4,000 reasoning tokens
Naive estimate / call
$0.035
ignores reasoning tokens
Vs naive estimate
3.9×
more than a plain token counter
Monthly cost
$135.00
1,000 calls/mo

What you pay for per call

Input
$0.017%
Reasoning
$0.1074%
Visible output
$0.02519%

Per-call cost breakdown

ComponentTokensRateCost
Input tokens2,000$5/1M$0.01
Reasoning tokens (billed as output)4,000$25/1M$0.10
Visible output tokens1,000$25/1M$0.025
True cost per call$0.135

Reasoning tokens add $100.00 per month at 1,000 calls, money a naive estimate never shows. Excludes prompt caching, batch discounts, and per-tenant contract pricing.

Cost curve of thinking harder

EffortReasoning tokensCost / callMonthly
Low1,000$0.06$60.00
Mediumselected4,000$0.135$135.00
High12,000$0.335$335.00
Max25,000$0.66$660.00

Same prompt and visible output; only the reasoning-token estimate changes. These are planning estimates, read your real reasoning_tokens usage value for exact figures.

Reasoning (“thinking”) tokens are billed at the model's output rate by Anthropic, OpenAI, and Google. Anthropic rates are cross-checked against the Anthropic pricing page; OpenAI and Gemini rows are published list prices that change without notice. Rates last verified 2026-06-06. Full sources are listed below the calculator.

How it works

Reasoning models think before they answer. That thinking is a stream of reasoning tokens (OpenAI's term), thinking tokens (Anthropic, Google) that the model generates internally. You usually never see them, the API discards or summarises the trace, but every provider counts and bills them, and crucially they are billed at the model's output rate, not the cheaper input rate.

A naive cost estimate prices only what you can point at: the prompt you sent and the answer you received. The real bill adds a third, invisible line. The calculator above uses this model:

  1. Convert the published per-million rates to per-token prices: Pin = input$/1M ÷ 1,000,000 and Pout = output$/1M ÷ 1,000,000.
  2. inputCost = inputTokens × Pin
  3. reasonCost = reasoningTokens × Pout , the hidden cost.
  4. outputCost = visibleTokens × Pout
  5. costPerCall = inputCost + reasonCost + outputCost, while naivePerCall = inputCost + outputCost. The multiplier is their ratio, and the monthly figure is costPerCall × callsPerMonth.

Because reasoning tokens and visible tokens are both billed at Pout, the per-call cost can also be written inputCost + (reasoning + visible) × Pout. The tool computes the cost both ways and they agree to the cent, a built-in cross-check on the math. Effort levels (Low, Medium, High, Max) seed a typical reasoning-token estimate; reasoning counts are non-deterministic, so the field is editable and the methodology is a planning aid, not a guarantee. For an exact figure, read your real reasoning_tokens value from the API response.

Worked examples

Claude Opus 4.8, coding agent, adaptive thinking

Input $5/1M, output $25/1M · I=2,000 R=8,000 V=1,000 · 1,000 calls/mo

  1. inputCost = 2,000 × $0.000005 = $0.010
  2. reasonCost = 8,000 × $0.000025 = $0.200
  3. outputCost = 1,000 × $0.000025 = $0.025
  4. costPerCall = $0.235 vs naive = $0.035
  5. multiplier = 0.235 ÷ 0.035 = 6.7×
  6. monthly = $0.235 × 1,000 = $235 (≈ Rs 71,675 at 305)

OpenAI o1, high reasoning effort

Input $15/1M, output $60/1M · I=1,500 R=25,000 V=500 · 1,000 calls/mo

  1. inputCost = 1,500 × $0.000015 = $0.0225
  2. reasonCost = 25,000 × $0.000060 = $1.500
  3. outputCost = 500 × $0.000060 = $0.030
  4. costPerCall = $1.5525 vs naive = $0.0525
  5. multiplier = 1.5525 ÷ 0.0525 = 29.6×
  6. monthly = $1.5525 × 1,000 = $1,552.50, the '30× my estimate' case

Edge case, reasoning with no visible output

Claude Sonnet 4.6, output $15/1M · I=0 R=5,000 V=0 · 1 call

  1. inputCost = 0
  2. reasonCost = 5,000 × $0.000015 = $0.075
  3. outputCost = 0
  4. costPerCall = $0.075 vs naive = $0.000
  5. A naive counter shows $0, but you are billed $0.075.
  6. The multiplier is infinite, so the tool shows 'n/a'.

Frequently asked questions

Sources & references

Anthropic rates were cross-checked against the Anthropic pricing page on 2026-06-06. OpenAI and Gemini rows are published list prices that change without notice, re-check them against the linked source before relying on a figure. No live API calls are made; pricing is a static, dated table.

Related tools

AI

AI Thinking Budget Calc

See the valid thinking-budget range each reasoning model accepts, Gemini 2.5 thinkingBudget, Claude budget_tokens, GPT-5/o-series reasoning_effort, then price the per-request and monthly cost of your setting, since thinking tokens are billed as output. Runs entirely in your browser.

Open tool

AI

AI API Cost Calculator

Estimate the monthly and per-request USD and LKR bill for any major LLM API. Pick a model, enter input/output tokens and requests per month, and compare every model cheapest-first, with optional 50% batch and cached-input discounts.

Open tool

AI

AI Reasoning Model Compare

Side-by-side comparison of the major AI reasoning models, OpenAI o3 & GPT-5 thinking, Claude extended thinking, Gemini thinking, DeepSeek-R1, Grok 4 and Qwen3, by reasoning-token pricing, context window, knowledge cutoff and AIME / GPQA / SWE-bench scores. Sort, filter, pick by use case, and estimate a task's true cost once hidden thinking tokens are counted. Sources cited, no signup.

Open tool

Comments & feedback

Spotted a bug or want an improvement? Tell us, our team reviews every comment, and good ideas get built. Comments are public and anonymous.

Found a bug, edge case, or want to suggest an improvement?

Email me at [email protected] , most fixes ship within 24 hours.