LLM API Cost Calculator
Enter your request volume below and see what the same workload costs on 9 models from DeepSeek, MiniMax, Alibaba, Zhipu, OpenAI and Anthropic — published prices with region and billing conditions. Add a cache-hit assumption, pick a baseline model to see savings, or edit any rate yourself. No sign-up.
Your estimated monthly cost per model
| Model | Provider | Base input $/1M | Base output $/1M | Cache $/1M | Input cost | Output cost | Total / month | vs baseline |
|---|
Current published rates, verified: November 9, 2026. Published API prices per 1M tokens — each row links to the provider's own pricing page below.
Estimates only, using the rates and conditions listed below. Peak share applies to DeepSeek; batch mode uses only verified batch rates. The cache ratio applies the provider's published cache-hit rate to that share of input tokens; cache-write premiums, storage and provider-specific cache eligibility rules are not modeled. OpenAI and Qwen long-input tiers are applied to the average input length; actual requests may vary. Taxes, cache writes, storage and tool fees are excluded. “Lowest listed” marks the lowest cost among the models listed here under your current assumptions, not a quality ranking. Always confirm on the provider's official pricing page.
Edit rates (custom quotes or changed prices)
Published rates used in this calculator
| Model | API name | Base input $/1M | Base output $/1M | Cache hit $/1M | Conditions & source |
|---|---|---|---|---|---|
| DeepSeek V4.1-Flash | deepseek-flash | 0.15 | 0.60 | 0.003 | Provider direct USD API; standard routing. Off-peak base prices. Peak input, cache reads and output cost 2×. Weekdays 01:00–04:00 and 06:00–10:00 UTC, excluding Chinese public holidays. Verified 2026-11-09. Official price ↗ |
| MiniMax M3.1-Flash | MiniMax-M3.1-Flash | 0.15 | 0.60 | 0.015 | Provider direct USD API; standard routing. Direct API standard rate; priority service excluded. Verified 2026-11-09. Official price ↗ |
| Qwen 3.7 Plus | qwen3.7-plus | 0.32 | 1.28 | — | Alibaba Cloud US (Virginia), US deployment. US deployment; current 20% discounted rate, up to 256K input. Above 256K: input/output ×3. Other regions differ; cache not modeled. Verified 2026-11-09. Official price ↗ |
| GLM-5 | glm-5 | 1.00 | 3.20 | 0.200 | Provider direct USD API; standard routing. Direct Z.ai API; cache storage fees excluded. Verified 2026-11-09. Official price ↗ |
| GPT-5.6 Luna | gpt-5.6-luna | 0.20 | 1.20 | 0.020 | Provider direct USD API; standard routing. Current official API rate. Above 272K input tokens: input ×2, output ×1.5; cache-write premiums excluded. Verified 2026-11-09. Official price ↗ |
| Claude Opus 5.5 | claude-opus-5-5 | 4.00 | 20.00 | 0.200 | Claude first-party API, global routing. Global routing; US-only residency and fast mode excluded. Batch is asynchronous; this calculator does not combine batch with cache discounts. Verified 2026-11-09. Official price ↗ |
| GPT-5.6 Terra | gpt-5.6-terra | 2.00 | 12.00 | 0.200 | Provider direct USD API; standard routing. Current official API rate. Above 272K input tokens: input ×2, output ×1.5; cache-write premiums excluded. Verified 2026-11-09. Official price ↗ |
| GPT-5.6 Sol | gpt-5.6-sol | 4.00 | 20.00 | 0.400 | Provider direct USD API; standard routing. Current official API rate. Above 272K input tokens: input ×2, output ×1.5; cache-write premiums excluded. Promotional rate through November 30, 2026; recheck before budgeting later months. Verified 2026-11-09. Official price ↗ |
| Claude Fable 5 | claude-fable-5 | 10.00 | 50.00 | 1.000 | Claude first-party API, global routing. Global routing; US-only residency and fast mode excluded. Batch is asynchronous; this calculator does not combine batch with cache discounts. Verified 2026-11-09. Official price ↗ |
Verified November 9, 2026 against each provider's official pricing page. See each rate’s conditions, region and effective date below. See also: full 2026 LLM pricing comparison · DeepSeek API pricing guide · OpenRouter vs calling models directly · data disclaimer
How the calculation works
LLM APIs charge per token. With a cached-input token fraction c (0 to 1), your monthly cost is:
(requests × input tokens × ((1−c) × input price + c × cache-hit price) + requests × output tokens × output price) ÷ 1,000,000
With c = 0 this reduces to the plain formula: (requests × input tokens × input price + requests × output tokens × output price) ÷ 1,000,000. Cache-write premiums and provider-specific cache eligibility are not modeled — treat results as a budgeting estimate.
A token is roughly 4 characters or 0.75 English words. A typical chat exchange uses about 1,000 input tokens (your prompt plus any context you attach) and 500 output tokens (the reply). Coding agents and RAG pipelines usually send far more input tokens — often 10,000+ per request — which is exactly when choosing a cheap-input model matters most.
Cutting your bill: three levers, and how to validate them
- Switch the model. Moving routine chat workloads from a frontier model to a budget open-weight model is often an order of magnitude cheaper. Don't take the price spread as proof of interchangeability — test accuracy, latency and failure rate on your own tasks first, then move traffic gradually.
- Cut input tokens. Trim system prompts, use prompt caching where offered, and send only relevant context. Input often dominates agentic workloads — set a cache-hit ratio above and watch the ranking change.
- Watch the middleman fee. Aggregators are convenient but charge on top of model rates — OpenRouter, for example, adds a 5.5% fee on credit purchases (8% on Business). At production volume, calling the provider directly — or a low-fee gateway — matters. We break this down in OpenRouter vs direct.
Frequently asked questions
How is LLM API cost calculated?
Per token. Monthly cost = (requests × input tokens × input price + requests × output tokens × output price) ÷ 1,000,000. Input tokens are what you send; output tokens are what you get back.
Which LLM API is the cheapest?
The lowest estimate depends on your token mix, cache ratio and billing mode. See the highlighted rows above. Cheapest is not the same as best for your task — validate quality on your own workloads before switching.
Does the calculator account for caching?
Optionally. Enter an estimated cache-hit ratio and models with a published cache-hit rate apply it to that share of input tokens (models without one keep the standard input rate). Cache-write premiums and provider-specific rules are not modeled — see the disclaimer.
Can I use custom quotes or updated prices?
Open the rate panel under the results and edit the input, output or cache rate — results update instantly, a reset button restores the official rates, and your overrides travel in the shareable link so your team sees the same assumptions.