Development · AI

AI API cost calculator

Compare token prices and monthly estimates for a stored snapshot of 69 AI models from 11 providers. Current prices and access may differ.

Last updated 2026.09.17 Price snapshot 2026-09-17 · 11 providers Processing method Local calculation in the browser Translation updated: 2026-11-02

tokens · usage · comparison

Compare AI API token costs

Enter usage and select models to compare estimated cost, value rankings and Toolvil’s reference performance scores.

Waiting

Enter input/output tokens per call and daily calls, then choose at least 2 models. Estimates use stored standard public API rates. Actual billing depends on options, account access and contract terms.

Select at least 2 models.

Input tokens per call (Input)

Output tokens per call (Output)

Calls per day

Exclude free models Show EUR equivalent All selected

Exchange rate and estimate scope

Stored USD rates: 2026-09-17. This estimate uses 30 days per month. Fees, taxes, discounts, long-context tiers and hidden reasoning usage are not applied automatically. Verify current pricing and model access.

US dollars per euro

ECB reference: 2026-10-30. 1 EUR = 1.1172 USD.

ECB exchange reference

Choose models (69 available)

Search models

Select at least 2 models to compare costs.

Cost formula [(input tokens ÷ 1M × input rate) + (output tokens ÷ 1M × output rate)] × calls/day × 30

Price snapshot Pricing snapshot: 2026-09-17, 11 providers · reference scores normalized by Toolvil

Note Batch, caching, long contexts, priority/flex, reasoning tokens, tools and taxes need separate billing checks.

Guide to the AI API cost calculator

This calculator helps product teams, developers and operators compare token-cost scenarios for 69 stored text and coding model rows from 11 providers. The underlying official-price snapshot was reviewed on 2026-09-17. It includes GPT-5.6 Sol, Terra and Luna, GPT-5.4 nano, Claude Opus 5 and Sonnet 5, Gemini 3.7 Flash, 3.6 Flash and 3.5 Flash-Lite, plus xAI Grok, DeepSeek V4, Qwen3 Coder, Kimi, GLM, MiniMax, Mistral and Amazon Nova. Enter input tokens, output tokens and calls per day to estimate a 30-day month for a prototype, support chatbot, coding agent or document-summary workflow. Availability and newer models may have changed since the snapshot.

An API bill involves more than a price table. Output can cost more than input, and internal reasoning or thinking tokens may add output charges. The same 1,000 input tokens and 500 output tokens can have very different costs across GPT-5.5, DeepSeek V4 Flash, Gemini Flash and Claude Opus rows. Compare stored standard-rate scenarios here, then check current official prices, token limits, regional terms and account availability before committing a budget. USD is the pricing basis; EUR amounts are illustrative conversions, not a separate European provider tariff.

Use cases

Compare monthly support-chatbot costs across stored GPT, Claude, Gemini and DeepSeek rows
Shortlist stored Kimi, Qwen, GLM, Mistral and Grok rows for coding-agent cost comparisons
Explore budgets for output-heavy document summaries, log analysis or assisted translation

Before you enter usage

Include prompts, system messages and RAG context in the input-token estimate.
Allow for generated answers, reasoning, JSON and code when estimating output tokens.
Estimate average daily traffic and batch workloads separately rather than treating a peak as the daily average.

How to use

Enter input tokens, output tokens and calls per day.
Choose whether to exclude free rows and show the alternate currency equivalent.
Search by a model or provider name such as GPT, Sonnet, DeepSeek or Qwen.
Use provider tabs and model cards to select at least 2 rows.
Compare monthly cost, reference scores and value rankings in summaries, bar charts, the scatter plot and sortable table.

Input valuesInput/output tokens, daily calls and selected models

Output valuesPer-call and monthly cost, USD/EUR equivalents, reference scores and value rankings

Model coverageStored snapshot of 69 text and coding models from 11 providers

Pricing and currency basisOfficial-price snapshot reviewed 2026-09-17; underlying rates in USD, with dated or custom USD/EUR conversion

Processing locationLocal browser calculation; usage inputs are not sent to a server

How to read the prices

Stored input/output rates are quoted per 1M tokens in USD. The snapshot records promotional Gemini 3.8, 3.7 and 3.6 Flash rates through 2026-12-31, with $1.50/$7.50 noted for 2027; GPT-5.6 Sol is recorded at $4/$20 with $5/$30 reference pricing and a promotion available at least through 2026-11-21. Confirm continuation or replacement with official sources. Standard or short-context tiers are used where providers have multiple context or processing tiers. Qwen uses international Singapore list prices. DeepSeek rows use peak cache-miss input/output rates; the stored off-peak note records a 50% reduction. MiniMax M3 has separate rows for up to 512K and more than 512K input tokens. These rows do not automatically switch tiers or apply discounts: read each note and check the actual account conditions.

Ways to reduce costs

Model routing: Evaluate less expensive models for classification, short summaries and preprocessing, reserving more capable models for complex reasoning or final answers.
Prompt caching: Check caching discounts for repeated system prompts, document context and tool definitions.
Output limits: When long answers are unnecessary, test a smaller output-token limit and a concise response format.
Batch processing: For non-urgent workloads, compare batch, flex or offline options against standard pricing.

Official pricing sources

Before budgeting, check the official pricing pages for OpenAI, Anthropic, Google, xAI, DeepSeek, Alibaba Cloud Model Studio, Kimi, Z.AI, MiniMax, Mistral and AWS Nova. Toolvil’s normalized reference scores aid relative comparisons; they do not guarantee an independent benchmark ranking or suitability for your workload. Rows without a reviewed score remain in cost comparisons but are excluded from score and value rankings. Evaluate your own tasks and confirm that each model is available to your account.

Calculation basis and sources

Calculation scope: Applies the stored standard input/output rates to your tokens per call and daily calls. Caching, batches, priority processing, long contexts, tools and taxes are not added automatically.
Calculation method: [(input tokens ÷ 1,000,000 × input rate) + (output tokens ÷ 1,000,000 × output rate)] × daily calls × 30 days. USD is the calculation basis; EUR uses the displayed exchange rate.
Limits and disclaimer: Illustrative estimates from a dated official-price snapshot. Check current prices and account, region, tier and contract conditions before budgeting. Unreviewed model scores are not invented.
Personal information processing: Usage inputs and calculations stay in your current browser; they are not sent to a server or stored.

🗓️ Source and copy review 2026-09-17 · 🔁 Next review due 2026-10-17

Reference sources

FAQ

How are AI API token costs calculated?

Divide input tokens by 1,000,000 and multiply by the input rate. Do the same for output tokens and add both amounts for cost per call. Multiply by daily calls and 30 for a monthly estimate. Rates are calculated in USD; EUR displays convert the same estimate.

What is the date of this price snapshot?

The underlying table records official pricing reviewed for all 11 providers on 2026-09-17. It is not a live feed or a complete current model directory. Prices, access, region, batching, caching, long-context tiers and priority options can change. Check the current official pages and the model available to your account before relying on the estimate.

Which GPT, Claude and Gemini rows are included?

The stored table includes GPT-6 Astra, GPT-5.6 Sol, Terra and Luna, Claude Fable 5.1, Opus 5 and Sonnet 5, Gemini 3.8 and 3.7 Flash, Grok 4.6, DeepSeek V4.1 Flash, Qwen3.8 Max, Kimi K3 and GLM-5.3 among its text and coding rows. This snapshot does not promise inclusion of every newer model. Rows without reviewed scores participate in cost comparisons but not performance or value rankings.

Is the listed rate enough for reasoning models?

Not always. Internal reasoning or thinking tokens can add billed output tokens. Actual cost can vary with task difficulty and reasoning settings. Include these tokens in your output estimate and check the provider’s billing definition; this calculator does not infer hidden token usage.

Will the displayed currency amount equal my bill?

No. USD estimates use stored USD token rates. EUR displays use a dated ECB reference or your custom USD-per-EUR rate. Card exchange rates, fees, taxes, cloud marketplace charges and enterprise terms can differ. The estimate is not a guaranteed charge.