Profession Calculators
Tech & ITPopular

LLM API Token Cost Estimator

Estimate monthly token costs using model pricing verified August 19, 2026, including GPT-5.6 Sol, Terra, and Luna; Claude Fable 5, Opus 5, and Sonnet 5; Gemini 3.7 Flash; and custom hosted open-model rates.

Share:
Workload & Model Configuration

Permanent pricing announced August 10, 2026 and verified in Anthropic Platform docs.

Presets use standard short-context rates. Cache-write premiums, long-context tiers, regional processing uplifts, grounding, and cache storage charges are not included.

Monthly & Annual Cost Summary
Enter your request volume and token counts, then click Calculate to view detailed cost projections.

Embed This Calculator on Your Website

Add this free calculator to your blog, website, or CMS with a simple copy-paste embed code.

Introduction

LLM pricing changed materially during 2026. A cost model built around GPT-4o, Claude 3.5 Sonnet, or Gemini 1.5 no longer represents the current market. As of August 19, 2026, OpenAI API pricing lists GPT-5.6 Sol at $5 per million short-context input tokens, $0.50 per million cached input tokens, and $30 per million output tokens. GPT-5.6 Terra is $2, $0.20, and $12 respectively. Anthropic Platform pricing lists Claude Fable 5 at $10 input, $1 cache hits, and $50 output; Claude Opus 5 at $5, $0.50, and $25; and Claude Sonnet 5 at $2, $0.20, and $10. Google Gemini API pricing lists Gemini 3.7 Flash at a promotional $0.75 input, $0.075 cached context, and $3.75 output through December 31, 2026. This calculator applies those current standard rates to a production request workload.

What This Calculator Does

Select a current model or enter custom rates for a hosted open model. Then enter prompt tokens, output tokens, requests per day, cache hit percentage, and active days per month. The calculator estimates fresh input cost, cached input cost, output cost, monthly spend, annual run rate, average cost per request, and cost per 1,000 requests. The comparison table runs the same workload through current OpenAI, Anthropic, and Google presets. Presets use standard short-context token rates. They do not include cache-write premiums, long-context pricing, Google grounding charges, Gemini cache storage, OpenAI regional processing uplifts, Batch discounts, Flex pricing, or Fast mode.

The Formula

Daily Cost = Fresh Input Cost + Cached Input Cost + Output Cost | Fresh Input Cost = Daily Requests x Prompt Tokens x (1 - Cache Hit Rate) / 1,000,000 x Input Rate | Cached Input Cost = Daily Requests x Prompt Tokens x Cache Hit Rate / 1,000,000 x Cached Input Rate | Output Cost = Daily Requests x Output Tokens / 1,000,000 x Output Rate | Monthly Cost = Daily Cost x Active Days

Providers publish token prices per one million tokens. Fresh input is prompt content processed at the base input rate. Cached input is eligible repeated context read from a provider cache. Output includes generated response tokens and, where the provider specifies it, billed thinking tokens. This calculator treats the entered cache percentage as cache hits. It does not calculate the separate cost to write a cache. Anthropic currently charges cache-write premiums, while Gemini charges context cache storage per token-hour. OpenAI also separates standard short-context and long-context rates for current GPT-5.6 models. Use custom rates or add those charges outside this result when they apply.

Step-by-Step Example

1

Choose a current model

Select Claude Sonnet 5 at $2 per million input tokens, $0.20 per million cache-hit tokens, and $10 per million output tokens. These rates were verified on August 19, 2026.

2

Enter request token volume

Enter 2,000 prompt tokens, 600 output tokens, and 5,000 requests per day. That workload processes 10 million prompt tokens and 3 million output tokens daily.

3

Enter the measured cache hit percentage

Enter a 40% cache hit rate only if provider usage reports support it. The daily prompt volume becomes 6 million fresh tokens and 4 million cached tokens.

4

Review the estimate

Claude Sonnet 5 costs $12 for fresh input, $0.80 for cached input, and $30 for output per day. Total daily cost is $42.80, monthly cost over 30 days is $1,284, and annual run rate is $15,408. Cache-write charges are not included.

Real-World Use Cases

Agentic application budgeting

An engineering team compares GPT-5.6 Sol, Claude Fable 5, Claude Sonnet 5, and Gemini 3.7 Flash before choosing separate models for planning, tool execution, and routine summarization.

Prompt-cache return analysis

A support platform measures how much repeated system instructions and product documentation reduce monthly read cost while separately tracking cache writes and storage.

Hosted open-model comparison

A team using an open-weights model enters its inference host's current input and output rates instead of relying on a stale generic Llama price.

Price change impact review

Finance reruns the same traffic assumptions when a provider changes standard pricing, promotional pricing, or model availability.

Comparison

ModelInput per 1MCached input per 1MOutput per 1MPricing note
OpenAI GPT-5.6 Sol$5.00$0.50$30.00Standard short-context
OpenAI GPT-5.6 Terra$2.00$0.20$12.00Standard short-context
OpenAI GPT-5.6 Luna$0.20$0.02$1.20Standard short-context
Anthropic Claude Fable 5$10.00$1.00$50.00Cache-hit rate; cache writes cost more
Anthropic Claude Opus 5$5.00$0.50$25.00Cache-hit rate; cache writes cost more
Anthropic Claude Sonnet 5$2.00$0.20$10.00Permanent pricing announced August 10, 2026
Google Gemini 3.7 Flash$0.75$0.075$3.75Promotional through December 31, 2026

Common Mistakes to Avoid

  • Using retired models as current benchmarks. Claude 3.5 Sonnet and Gemini 1.5 are historical comparison points, not suitable default presets for an August 2026 production budget.

  • Treating cache hits and cache writes as the same charge. Anthropic has separate 5-minute and 1-hour cache-write prices. Google bills cached context reads and token-hour storage separately.

  • Ignoring long-context tiers. OpenAI GPT-5.6 standard pricing is higher for long-context requests. A short-context preset understates workloads that cross the provider threshold.

  • Leaving thinking tokens out of output. Gemini 3.7 Flash states that billed output includes thinking tokens. Reasoning workloads can therefore produce higher billed output than the visible answer.

  • Assigning one generic price to open-source models. Open weights do not have one universal API rate. Hosted pricing varies by provider, hardware, quantization, batching, and service level.

Frequently Asked Questions

Which models are current in this calculator on August 19, 2026?

The presets include GPT-5.6 Sol, Terra, and Luna; GPT-5.4 Mini; Claude Fable 5, Opus 5, Sonnet 5, and Haiku 4.5; and Gemini 3.7 Flash. Custom pricing covers hosted open models and other providers.

Does the calculator include Anthropic cache-write charges?

No. It models fresh input and cache hits. Anthropic lists 5-minute and 1-hour cache-write rates separately, so add those costs when your application creates or refreshes cached prefixes.

Why does Gemini 3.7 Flash have a date-sensitive price?

Google lists $0.75 input, $3.75 output, and $0.075 cached context through December 31, 2026, with higher rates beginning January 1, 2027. Recalculate after the promotional period.

Does this include web search or grounding charges?

No. Google Search grounding, Google Maps grounding, OpenAI tool calls, and other feature charges are separate from text token rates.

How should I price a self-hosted open model?

Convert GPU rental, idle capacity, orchestration, storage, and engineering overhead into effective input and output rates per million tokens, then enter those custom rates.

Accuracy and Disclaimer

Pricing was researched from official OpenAI, Anthropic, and Google developer documentation and verified on August 19, 2026. Provider prices, model availability, promotional periods, context thresholds, and feature charges can change without notice. The calculator is a planning estimate, not an invoice. Confirm current rates on the linked official pricing pages before purchasing capacity or setting customer prices.

Conclusion

Token rates are only one part of production AI cost. Record the model identifier, service tier, context length, cache-write volume, cache-hit volume, and tool charges from your real invoices. Recheck published pricing before approving a budget because providers can change rates or retire models. After estimating token spend, use the Vector Database and Embedding Storage Sizer for retrieval infrastructure and the Cloud Egress and Data Transfer Cost Estimator for network costs.