AI Model API Cost Calculator — Token & Image Budgets

Model ID deepseek-flash, checked 2026-10-09. Choose Peak or Off-peak manually and enter an input cache hit percentage. Peak USD/MTok: cache hit $0.006, cache miss $0.30, output $1.20; Off-peak: $0.003, $0.15, $0.60. The input cost weights cache-hit and cache-miss rates; all billable output including thinking tokens uses the output rate, then calls multiply the total. Cache hits are an assumption. This tool does not detect holidays or current peak hours, call models, or rank model quality. Fable 5.1 also models warm-cache reads at its own verified rate; other models retain uncached rates; third-party proxy markups and other fees are excluded.

Official Google API pricing checked 2026-10-07: model ID gemini-nano-banana-2.1. Standard image output costs $0.0336 at 1K, $0.0504 at 2K and $0.1134 at 4K from 1,120 / 1,680 / 3,780 tokens at $30 per million image tokens. Batch uses $15 per million image tokens: $0.0168 / $0.0252 / $0.0567. The official Standard 4K display rounds to $0.113; this calculator retains token-based precision. Batch is asynchronous and separate from Standard. Default budgets cover image output only; inputs, text/thinking output and chargeable Standard searches require your own totals. Batch grounding is excluded because its pricing is not listed. The API free tier is not available, and we do not read your account balance or generate images.

Cost = (input tokens ÷ 1,000,000 × input rate + output tokens ÷ 1,000,000 × output rate) × number of calls. For the Claude models, the rates are Anthropic’s official base standard rates in USD per million tokens, verified 2026-09-30. The Gemini 4 Argon entry uses Google’s announced introductory rates, checked 2026-10-02, and the Qwen3.5-Flash entry uses Alibaba Model Studio’s standard Singapore / International rates, checked 2026-10-05.

The calculator lists Anthropic’s current-generation lineup as of 2026-09-30: Claude Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 4.5, with their official model IDs. Haiku 5.5 was added with prompt-length tiers checked October 8, 2026. It also includes a Gemini 4 Argon entry at Google’s announced introductory rate, a Qwen3.5-Flash entry at Alibaba Model Studio’s standard Singapore / International rate, and a GPT-5.6 Sol entry at OpenAI’s promotional API rate. Older legacy models are not listed because their current rates were not re-verified at the data date; the same usage can be compared across all included models side by side.

DeepSeek V4.1 Flash explicitly models manually chosen peak/off-peak rates and the input cache hit percentage. Fable 5.1 also uses the hit percentage for already-warmed eligible cache reads at $0.25/MTok, with $10 uncached input and $50 output (Claude API Standard global routing, checked October 14, 2026). Initial/cache writes, expiry and minimum requirements, Batch/Fast/US-only tiers, tools, third-party fees, tax and subscriptions are excluded; the subtotal difference is not guaranteed savings or a complete bill. Other token models use uncached standard rates; GPT-5.6 Sol applies its published per-request long-context surcharge, and Haiku 5.5 applies its published 100,000-prompt-token tiers. The separate Nano Banana image budget offers Standard and asynchronous Batch pricing. Other caching, batch, long-context and priority tiers are not modeled. Unknown usage and other fees can change the actual bill.

No. The estimates are third-party developer API token costs. They are not Claude or ChatGPT subscription fees, and this website is free to use—these costs are not related to this site.

Checked 2026-10-05 at support.claude.com: the Pro web plan is $20 per month, Max 5x is $100 per month, and Max 20x is $200 per month (web USD before tax; Max is billed monthly). Max raises session and weekly usage limits relative to Pro—the official pages do not promise a fixed message or token count. The #claude-max comparison above shows which plans fit a monthly budget and the full-month difference; subscriptions and API billing stay separate.

Checked 2026-10-06 on OpenAI’s official model page: the promotional API rate is $4 per million input tokens and $20 per million output tokens (model ID gpt-5.6-sol; gpt-5.6 is an alias of the same model). Promotional rates are guaranteed at least through November 21, 2026 — that date is not a promise of a later price change, and the official page must be rechecked after it. Above 272,000 input tokens per request, the full request uses 2× input and 1.5× output token rates; 272,000 exactly stays at the base rates. Cached tokens, cache writes, batch discounts, tool fees, taxes and hosted subscriptions are not included.

No. The calculator runs entirely in your browser with published rates. It never calls a model API, and nothing you type is sent to a server or stored.

Verified 2026-09-30 from platform.claude.com: Fable 5.1 costs $10 per million input tokens and $50 per million output tokens; Opus 5.5 costs $4/$20; Sonnet 5.5 costs $2/$10; Haiku 4.5 costs $1/$5.

Gemini 4 Argon is listed at Google’s announced introductory rate of $2 per million input tokens and $10 per million output tokens, announced September 30, 2026 and checked 2026-10-02. Public access is not confirmed and Google has not announced a model ID, so this entry is a token budget estimate only—it does not provide API access, and caching, batch discounts, long-context pricing, taxes, and subscription fees are excluded.

The Qwen3.5-Flash entry uses Alibaba Model Studio’s standard rates for the Singapore / International scope, checked 2026-10-05: $0.10 per million input tokens and $0.40 per million output tokens, with model ID qwen3.5-flash. Cost = (input tokens ÷ 1,000,000 × $0.10 + billable output tokens ÷ 1,000,000 × $0.40) × number of calls. Only those standard input and output rates are modeled.

Yes. When a request uses Qwen3.5-Flash’s hybrid thinking, the thinking (chain-of-thought) tokens are billed as output tokens. Enter the total billable output tokens, including thinking tokens when applicable, so the estimate matches the provider’s bill.

No. The entry reflects Alibaba Model Studio’s standard Singapore / International scope only. Rates for other regions and scopes—including China and US (Global) tiered pricing—can differ, and reseller or consumer-subscription prices are not the same as these API rates. Explicit cache creation/read, batch pricing, promotional offers, taxes, and extra tool fees are also excluded, so confirm current rates at the official source shown with the calculator.

Standard API rates checked October 8, 2026: up to 100,000 prompt tokens per request costs $0.10 input / $0.50 output per million tokens; above 100,000 costs $0.50 / $2.50 for the whole request. Exactly 100,000 stays at the lower tier. Calls multiply the budget after tier selection. Enter actual billable prompt tokens including system instructions, history and tool schemas. Caching, Batch, tool fees, taxes, subscriptions and regional premiums are excluded. See the linked official Anthropic pricing and model overview sources.

One model, mistral-large-4, with two standard uncached token price scenarios: default list price $1.36 input / $4.18 output per million tokens, or launch sale $0.68 / $2.09. Checked 2026-10-08. The official October 6, 2026 changelog says 50% off for two weeks; the exact cutoff is unspecified. Recheck the provider before purchase. Comparison and copy use the selected scenario. Public Preview; this calculator does not provide access. Cached input, batch, priority, tools, third-party fees, taxes and subscriptions are excluded.