Cost per Model Instance Calculation Tool for LLMs (Large Language Models)
Calculate the cost for single LLM output across 12 cloud providers and Ollama on-premises. Compare costs with 7 types of outputs, using both prompt caching and Batch APIs in Series FinOps AI #33.
Disclaimer: prices are subject to change as of May 2026 (manual snapshots). Default token value is average estimates. Verify official price lists before making a purchase decision.
Configure Outputs and Parameters
Select Templates
Selected items5Cost Comparison per Unit of Production
Chat Reply| Model/ Provider | Cost/output | Daily | Monthly (30 days) | Savings vs First Place |
|---|---|---|---|---|
| Gemini 2.5 FlashGoogleCheaper | $0.00017 | $0.17 | $5.10 | -88% |
| DeepSeek-V3DeepSeek | $0.00070 | $0.70 | $21.00 | -50% |
| GPT-5 miniOpenAI | $0.0014 | $1.40 | $42.00 | Benchmarking |
| Claude Haiku 4.5Anthropic | $0.0016 | $1.60 | $48.00 | +14% |
| Ollama llama3.1:8b (on-prem)Meta / Self-hostedSelf-Hosted (Free) | Free on premises version available, contact Federico Calò for details. | $0.00 | $0.00 | -100% |
Monthly Cost Per Model
(on-premise Ollama excluded for scale)Come utilizzare LLM Cost Per Output Calculator
Choose output type
Select one of the 7 types (chat reply, RAG, summarization, code generation, agent loop, embeddings, classification): the tool estimates average input/output tokens for that type.
Set Volume and Optimizations
Indicate daily requests, any manual token overrides and enable prompt caching and/or batch API if used in production.
Compare selected models
Select models to compare among 12+ cloud providers + On-Prem and read output cost, daily/ monthly cost and savings percentage compared to the baseline.
Suggerimenti
- Use preloaded "RAG" and "Agent Loop" scenarios to quickly see how recommended models change based on the type of output.
- If your workflow has repeated system prompts on every request, activate prompt caching to see the real impact on cost.
- Compare at least one premium cloud model and a more affordable one to understand the quality/price trade-off before choosing.
Domande frequenti
How are tokens valued if I don't insert them manually?
Every type of output has a default estimate calibrated for typical use cases (e.g., an agent loop uses many more tokens than a chat reply): you can always override input and output tokens with your actual values.
What changes when you enable prompt caching?
The prompt caching applies an estimated saving of 90% on the portion of input tokens that the provider considers a cache hit (e.g., repeated system prompts): the tool shows this as a cost reduction for the output.
Is there a batch API discount?
Batch API applies an estimated discount of 50% on the cost, typical of providers offering asynchronous processing that is not in real-time: useful for non-interactive workloads such as mass summarization or classification.
What's the difference with LLM Cost Optimizer?
LLM Cost Optimizer calculates the total cost of ownership (TCO) monthly/annually with break-even vs Ollama on-prem; this tool focuses on unitary cost per single output, useful for comparing providers at equal usage case.
Is the on-premises Ollama model always free for output?
In comparison, the local model is marked as such and does not enter into the monthly cost calculation for cloud providers' consumption: its real cost is that of the infrastructure hosting it, not estimated by individual output in this tool.