LLM cost calculator. Frontier vs open models.
Pick the model you pay for today and describe one request. See the monthly bill next to the same workload on open models.
Already on OpenAI or Anthropic? Import a usage export to fill this in.
CSV or JSON, or drop it here. Read in your browser, never uploaded.
Where to find it
- OpenAI: export the Usage page as CSV, grouped by model, or save the JSON from the Usage API (
/v1/organization/usage/completionswithgroup_by=model). - Anthropic: the tokens CSV from the Claude Console usage export, or the JSON from the Admin API usage report (
/v1/organizations/usage_report/messagesgrouped by model).
Which model do you use today?
Pick the closest one. Prices are each vendor's published list price.
What does one request do?
Pick the example closest to yours. It fills in typical sizes you can change below.
How many requests a month?
Every call your app makes counts once: one chat reply, one record checked, one document read.
About 33 a day.
Fine-tune (optional)20,000 in · 1,500 out · 80% cached · realtime
Everything you send: instructions, context, history. That is about 15,000 words, 30 pages.
The answer, plus any reasoning the model writes first. That is about 1,130 words, 2 pages.
The same system prompt, document or chat history sent again bills at the cheaper cached rate. Leave it at 0% if every prompt is different.
Our pick for agent step
Claude Sonnet 5 today
$26.20/mo
$0.03 per request
On DeepSeek V4.1-Flash
$2.35/mo
$0.0023 per request
Estimates at per-token rates. Steady traffic can run at one flat monthly rate instead.
You save
$286.25/yr
$23.85 a month, 91% less
Why DeepSeek V4.1-Flash? The open model we'd try first in place of Claude Sonnet 5 for agent work. The lowest price here is GLM 5.3 Flash at $1.00/mo. Try it on a sample of your own requests before you switch.
| Open model | Monthly | Saving |
|---|---|---|
| GLM 5.3 FlashZhipu | $1.00 | $25.2026x cheaper |
| DeepSeek V4.1-FlashDeepSeekOur pick | $2.35 | $23.8511x cheaper |
| GLM 5.3Zhipu | $5.81 | $20.394.5x cheaper |
| Qwen3.8 27BAlibaba | $7.54 | $18.663.5x cheaper |
| Gemma 4 31BGoogle | $22.04 | $4.16−16% |
Open models priced at OpenRelay's live realtime per-token rates. Cache writes excluded.
Every rate, per 1M tokens.
Frontier list prices as published on October 11, 2026. Open-model rates are OpenRelay's current prices.
Full inference pricing| Model | Input | Cached input | Output | Batch in / out |
|---|---|---|---|---|
| Claude Fable 5.1Anthropic | $10.00 | $0.25 | $50.00 | $5.00 / $25.00 |
| Claude Opus 5.5Anthropic | $4.00 | $0.20 | $20.00 | $2.00 / $10.00 |
| Claude Sonnet 5.5Anthropic | $2.00 | $0.10 | $10.00 | $1.00 / $5.00 |
| Claude Sonnet 5Anthropic | $2.00 | $0.20 | $10.00 | $1.00 / $5.00 |
| Claude Haiku 5.5AnthropicPrompts up to 100K tokens; longer prompts bill at 5x. | $0.10 | $0.01 | $0.50 | $0.05 / $0.25 |
| GPT-6 AstraOpenAI | $10.00 | $1.00 | $50.00 | $5.00 / $25.00 |
| GPT-6.1 SolOpenAI | $2.00 | $0.10 | $10.00 | $1.00 / $5.00 |
| GPT-5.5OpenAI | $5.00 | $0.50 | $30.00 | $2.50 / $15.00 |
| GPT-6 LunaOpenAI | $0.10 | $0.01 | $0.50 | $0.05 / $0.25 |
| Gemini 3.1 Pro PreviewGoogle | $2.00 | $0.20 | $12.00 | $1.00 / $6.00 |
| Gemini 3.8 FlashGoogleGoogle doubles these rates on January 1, 2027. | $0.75 | $0.075 | $3.75 | $0.375 / $1.875 |
| GLM 5.3 FlashZhipu · open weights on OpenRelay | $0.045 | $0.009 | $0.45 | Realtime used |
| Qwen3.8 27BAlibaba · open weights on OpenRelay | $0.42 | $0.085 | $3.00 | Realtime used |
| DeepSeek V4.1-FlashDeepSeek · open weights on OpenRelay | $0.15 | $0.006 | $1.10 | Realtime used |
| Gemma 4 31BGoogle · open weights on OpenRelay | $0.99 | $0.99 | $1.49 | Realtime used |
| GLM 5.3Zhipu · open weights on OpenRelay | $0.079 | $0.062 | $3.00 | Realtime used |
Price is half the decision.
Send us a sample of requests where you already trust the answers. We run them on open models and show the results next to your current model's.
Contact usQuestions
How does the calculator estimate cost?
Monthly cost is requests per month times (fresh input tokens x input rate + cached input tokens x cached rate + output tokens x output rate). Frontier rates are each vendor's published list price as of October 11, 2026; open-model rates are OpenRelay's live prices. For example, a million record-matching calls a month at 2,500 input and 150 output tokens cost $6,500 on Claude Sonnet 5 ($3,250 on its batch API) and $180.00 on GLM 5.3 Flash.
Is an open model as good as Claude, GPT or Gemini for my task?
It depends on the task, and price tables cannot answer it. Extraction, classification, matching and most RAG work move to open models with little loss; long multi-step reasoning is where the gap shows. Run a few hundred of your own requests with known good answers through both models and compare.
Do reasoning tokens count as output?
Yes. Every vendor bills thinking or reasoning tokens at the output rate, on frontier and open models alike. If your model reasons before answering, put the total generated tokens in the output field, not just the visible answer.
Are token counts the same across models?
No. Each model family has its own tokenizer, so the same text can be a different number of tokens. Anthropic states that Claude 4.7 and later models produce about 30% more tokens for the same text than earlier Claude models. Measure on the model you plan to use for an exact figure.
How do I cut the cost further?
Cache shared prompt prefixes, and send fewer tokens to the LLM. For matching and search, embed your records first (BGE-M3 on OpenRelay) and send only the top few candidates to the model instead of the whole set.
How hard is it to switch from Claude or GPT?
Open models on OpenRelay take OpenAI chat completions requests, and DeepSeek V4.1-Flash also serves the Anthropic Messages API, so switching is usually a base URL, an API key and a model name. Pay-as-you-go billing is per token with no subscription, and reserved throughput is sold at one flat monthly rate.
Run it on open models.
Per-token billing, no subscription, no contract.