LLM API Cost Calculator
Enter your usage profile and instantly compare what every major LLM provider would charge — per request and per month. The default scenario below is real, server-rendered output; change any field to recalculate live.
Estimated cost — per month
Shown in USD. Pricing last verified 2026-09-15.
API price = the provider's list price per 1M tokens (input → output). Your cost = estimate for the usage profile you set on the left (4,000 in + 1,000 out per request).
Tip: click a column header to sort, or filter by category above. Sorted by cost per request by default.
| Provider / Model | Context | API price / 1M in → out | Your cost / request | Your cost / month | Links |
|---|---|---|---|---|---|
| Mistral Mistral Nemo fast cheapest batch | 131K | $0.0190 → $0.0300 | $0.000106 | $0.1060 | Visit |
| IBM Granite Granite 4.0 Micro open-weight | 131K | $0.0170 → $0.1120 | $0.000180 ×1.7 | $0.1800 | Visit Review |
| OpenAI gpt-oss-20b fast cache batch | 131K | $0.0300 → $0.1300 | $0.000250 ×2.4 | $0.2500 | Visit Review |
| Qwen Qwen3.7 Flash open-weight cache | 1M | $0.0300 → $0.1300 | $0.000250 ×2.4 | $0.2500 | Visit Review |
| Amazon Nova Micro 1.0 fast | 128K | $0.0350 → $0.1400 | $0.000280 ×2.6 | $0.2800 | Visit |
| Meta Llama 3.1 8B Instruct open-weight cache | 131K | $0.0500 → $0.0800 | $0.000280 ×2.6 | $0.2800 | Visit Review |
| Mistral Mistral Small 3 fast batch | 33K | $0.0500 → $0.0800 | $0.000280 ×2.6 | $0.2800 | Visit |
| Cohere Command R7B (12-2024) fast | 128K | $0.0375 → $0.1500 | $0.000300 ×2.8 | $0.3000 | Visit |
| Google Gemma 3 4B fast batch | 131K | $0.0500 → $0.1000 | $0.000300 ×2.8 | $0.3000 | Visit |
| Meta Llama 3.2 1B Instruct open-weight | 60K | $0.0270 → $0.2010 | $0.000309 ×2.9 | $0.3090 | Visit Review |
| OpenAI gpt-oss-120b fast batch | 131K | $0.0370 → $0.1700 | $0.000318 ×3.0 | $0.3180 | Visit Review |
| Google Gemma 3 12B fast batch | 131K | $0.0500 → $0.1500 | $0.000350 ×3.3 | $0.3500 | Visit |
| Tencent Hy-MT2-1.8B open-weight | 8K | $0.0440 → $0.1770 | $0.000353 ×3.3 | $0.3530 | Visit Review |
| DeepSeek DeepSeek V4 Flash 0731 open-weight cache | 1.3M | $0.0600 → $0.1200 | $0.000360 ×3.4 | $0.3600 | Visit Review |
| Qwen Qwen3 30B A3B Instruct 2507 open-weight | 262K | $0.0482 → $0.1931 | $0.000386 ×3.6 | $0.3859 | Visit Review |
| NVIDIA Nemotron 3 Nano 30B A3B open-weight cache | 262K | $0.0500 → $0.2000 | $0.000400 ×3.8 | $0.4000 | Visit Review |
| Microsoft Phi 4 open-weight | 16K | $0.0700 → $0.1400 | $0.000420 ×4.0 | $0.4200 | Visit Review |
| Amazon Nova Lite 1.0 fast | 300K | $0.0600 → $0.2400 | $0.000480 ×4.5 | $0.4800 | Visit |
| IBM Granite Granite 4.2 8B open-weight cache | 131K | $0.0600 → $0.2500 | $0.000490 ×4.6 | $0.4900 | Visit Review |
| Mistral Ministral 3 3B 2512 fast cache batch | 131K | $0.1000 → $0.1000 | $0.000500 ×4.7 | $0.5000 | Visit |
| Mistral Mistral Small 3.2 24B fast batch | 256K | $0.0750 → $0.2000 | $0.000500 ×4.7 | $0.5000 | Visit |
| Self-hosted Open-weight on your GPU (illustrative) open-weight | 128K | $0.1000 → $0.1000 | $0.000500 ×4.7 | $0.5000 | Visit Review |
| NVIDIA Nemotron 3.5 Lightning open-weight cache | 262K | $0.0800 → $0.2000 | $0.000520 ×4.9 | $0.5200 | Visit Review |
| Qwen Qwen3.5-Flash open-weight | 1M | $0.0650 → $0.2600 | $0.000520 ×4.9 | $0.5200 | Visit Review |
| Meta Llama 3.2 3B Instruct open-weight | 131K | $0.0500 → $0.3300 | $0.000530 ×5.0 | $0.5300 | Visit Review |
| DeepSeek DeepSeek V4 Flash 0423 open-weight cache | 1M | $0.0886 → $0.1772 | $0.000532 ×5.0 | $0.5316 | Visit Review |
| Qwen Qwen3.5-9B open-weight | 262K | $0.1000 → $0.1500 | $0.000550 ×5.2 | $0.5500 | Visit Review |
| Qwen Qwen3 Coder 30B A3B Instruct open-weight | 262K | $0.0700 → $0.2800 | $0.000560 ×5.3 | $0.5600 | Visit Review |
| Tencent Hy-MT2-30B-A3B open-weight | 8K | $0.0740 → $0.2950 | $0.000591 ×5.6 | $0.5910 | Visit Review |
| Tencent Hy-MT2-7B open-weight | 8K | $0.0740 → $0.2950 | $0.000591 ×5.6 | $0.5910 | Visit Review |
| OpenAI gpt-oss-safeguard-20b fast cache batch | 131K | $0.0750 → $0.3000 | $0.000600 ×5.7 | $0.6000 | Visit Review |
| OpenAI GPT-5 Nano fast cache batch | 400K | $0.0500 → $0.4000 | $0.000600 ×5.7 | $0.6000 | Visit Review |
| Qwen Qwen2.5 7B Instruct open-weight | 33K | $0.1000 → $0.2000 | $0.000600 ×5.7 | $0.6000 | Visit Review |
| Qwen Qwen3 32B open-weight | 131K | $0.0800 → $0.2800 | $0.000600 ×5.7 | $0.6000 | Visit Review |
| Z.AI GLM 4.7 Flash open-weight | 200K | $0.0605 → $0.4000 | $0.000642 ×6.1 | $0.6420 | Visit Review |
| Google Gemma 4 26B A4B fast cache batch | 262K | $0.0900 → $0.3000 | $0.000660 ×6.2 | $0.6600 | Visit |
| Google Gemma 4 31B fast cache batch | 262K | $0.0900 → $0.3400 | $0.000700 ×6.6 | $0.7000 | Visit |
| Meta Llama 4 Scout open-weight | 1.3M | $0.1000 → $0.3000 | $0.000700 ×6.6 | $0.7000 | Visit Review |
| Mistral Voxtral Small 24B 2507 fast cache batch | 33K | $0.1000 → $0.3000 | $0.000700 ×6.6 | $0.7000 | Visit |
| Qwen Qwen3 235B A22B Instruct 2507 open-weight cache | 262K | $0.0875 → $0.3500 | $0.000700 ×6.6 | $0.7000 | Visit Review |
| Meta Llama 3.3 70B Instruct open-weight | 131K | $0.1000 → $0.3200 | $0.000720 ×6.8 | $0.7200 | Visit Review |
| Qwen Qwen3 14B open-weight | 131K | $0.1200 → $0.2400 | $0.000720 ×6.8 | $0.7200 | Visit Review |
| Mistral Ministral 3 8B 2512 fast cache batch | 262K | $0.1500 → $0.1500 | $0.000750 ×7.1 | $0.7500 | Visit |
| Google Gemma 3 27B fast cache batch | 131K | $0.0800 → $0.4500 | $0.000770 ×7.3 | $0.7700 | Visit |
| NVIDIA Nemotron 3 Super open-weight | 262K | $0.0800 → $0.4500 | $0.000770 ×7.3 | $0.7700 | Visit Review |
| Google Gemini 2.5 Flash Lite fast cache batch | 1M | $0.1000 → $0.4000 | $0.000800 ×7.5 | $0.8000 | Visit |
| OpenAI GPT-4.1 Nano fast cache batch | 1M | $0.1000 → $0.4000 | $0.000800 ×7.5 | $0.8000 | Visit Review |
| Qwen Qwen3 VL 32B Instruct open-weight | 131K | $0.1040 → $0.4160 | $0.000832 ×7.8 | $0.8320 | Visit Review |
| Meta Llama Guard 4 12B open-weight | 164K | $0.1800 → $0.1800 | $0.000900 ×8.5 | $0.9000 | Visit Review |
| Qwen Qwen3 8B open-weight | 131K | $0.1170 → $0.4550 | $0.000923 ×8.7 | $0.9230 | Visit Review |
Estimates only. Actual bills depend on exact token counts, tier pricing and provider changes. Always confirm on the provider's pricing page.
How LLM API pricing works
Every major LLM provider bills by the token — a chunk of text roughly ¾ of a word in English. You pay separately for input tokens (everything you send: system prompt, retrieved context and the user message) and output tokens (what the model writes back). Output is typically priced two to five times higher than input, which is why concise responses save real money at scale.
The formula
cost_per_request = (input_tokens / 1,000,000) × input_price_per_M
+ (output_tokens / 1,000,000) × output_price_per_M
cost_per_period = cost_per_request × requests_in_periodWhat moves the number
- Prompt caching. If you reuse a long prefix (a big system prompt, a fixed knowledge base), many providers bill those cached tokens at a fraction of the normal input price.
- Batch mode. Non-interactive workloads submitted in bulk often get a flat discount (commonly 0.5×) in exchange for slower, asynchronous delivery.
- Context tiers. Some providers raise the per-token price once a request exceeds a context threshold. The dataset supports tiered pricing for exactly this case.
- Self-hosting. Open-weight models on your own GPUs change the equation entirely: you trade per-token fees for fixed infrastructure cost and operational work. See our self-hosted cost breakdown.
Frequently asked questions
How are LLM API costs calculated?
Providers bill per million tokens, separately for input (your prompt) and output (the model's response). Cost per request = (input tokens / 1,000,000 × input price) + (output tokens / 1,000,000 × output price). Multiply by your request volume for the period.
What is the difference between input and output tokens?
Input tokens are everything you send to the model — system prompt, context and user message. Output tokens are what the model generates. Output is usually priced several times higher than input, so response length matters a lot for cost.
Does prompt caching reduce cost?
Yes, where supported. Repeated prompt prefixes (e.g. a long system prompt) can be billed at a steep discount. The calculator lets you set what share of input tokens are cached.
Why do prices vary so much between providers?
Model size, hardware efficiency, context window, and business strategy all play a part. Frontier reasoning models cost the most; fast/cheap and open-weight self-hosted options can be orders of magnitude cheaper for suitable tasks.