AI Model Fine-Tuning Cost Calculator
Full version ↗
Model to Fine-Tune
OpenAI — GPT-4o fine-tune (managed, $25/M train tokens)
OpenAI — GPT-4o mini fine-tune (managed, $3/M)
OpenAI — GPT-3.5 Turbo fine-tune (managed, $8/M)
Anthropic — Claude via AWS Bedrock (enterprise)
Google — Gemini 1.5 Flash fine-tune (managed)
Meta — Llama 3.3 70B self-hosted (A100 GPU)
Meta — Llama 3.1 8B self-hosted (budget GPU)
Meta — Llama 3.1 405B self-hosted (multi-GPU)
Mistral — Mistral 7B self-hosted
Mistral — Mistral Large via managed API
Microsoft — Phi-3 Mini self-hosted (very cheap)
Microsoft — Phi-3 Medium self-hosted
Google — Gemma 2 9B self-hosted
Alibaba — Qwen 2.5 7B self-hosted
Alibaba — Qwen 2.5 72B self-hosted
DeepSeek — R1 Distill 8B self-hosted
DeepSeek — R1 Distill 70B self-hosted
Cohere — Command R fine-tune (managed)
Training Data (thousands of tokens)
Training Epochs
Monthly Inference (millions of tokens)
Calculate
—
One-Time Training Cost
Monthly Inference Cost
—
Year 1 Total Cost
—
vs GPT-4o API (Year 1)
—
Powered by CalcWolf — Free Calculators