LLM Fine-Tuning Cost Calculator
Training cost of a fine-tuning job on Together AI, Fireworks, Vertex AI and OpenAI
Your training job
6,000,000 training tokens per run (dataset x epochs).
Training cost by provider
| Provider and model | Per 1M tokens | Per run | All runs |
|---|---|---|---|
| Together AI: Llama 3.1 8B LoRA SFT. $4 minimum per job. | $0.34 | $4.00 | $4.00 |
| Together AI: Qwen3.5 27B LoRA SFT. $4 minimum per job. | $1.05 | $6.30 | $6.30 |
| Together AI: Llama 3.3 70B LoRA SFT. $4 minimum per job. | $2.03 | $12.18 | $12.18 |
| Together AI: gpt-oss-120B LoRA SFT. $6 minimum per job. | $2.50 | $15.00 | $15.00 |
| Fireworks AI: Base model up to 16B Managed LoRA SFT | $0.50 | $3.00 | $3.00 |
| Fireworks AI: Base model 16.1B to 80B Managed LoRA SFT | $3.00 | $18.00 | $18.00 |
| Fireworks AI: Base model 80B to 300B Managed LoRA SFT | $6.00 | $36.00 | $36.00 |
| Google Vertex AI: Gemini 3.5 Flash Lite Supervised fine-tuning. Listed as $0.003 per 1,000 tokens. | $3.00 | $18.00 | $18.00 |
| Google Vertex AI: Gemini 3.5 Flash Supervised fine-tuning. Listed as $0.01 per 1,000 tokens. | $10.00 | $60.00 | $60.00 |
| OpenAI: gpt-4.1-mini Fine-tuning. Closed to new users; existing users only. | $5.00 | $30.00 | $30.00 |
| OpenAI: gpt-4.1 Fine-tuning. Closed to new users; existing users only. | $25.00 | $150 | $150 |
The formulas
Training tokens = dataset tokens x epochs (+ validation tokens x evaluations, Together only) Cost per run = max(minimum charge, training tokens / 1,000,000 x price per 1M) All runs = cost per run x training runs
How this works
Every provider here bills supervised fine-tuning per training token. Together AI, Fireworks and Google all define training tokens the same way: the tokens in your dataset times the number of epochs. Together also bills validation tokens times the number of evaluations, and has a minimum charge per job.
Prices are the published per-token rates for LoRA or supervised fine-tuning, read on November 11, 2026. Vertex AI lists its Gemini tuning price per 1,000 tokens, so $0.01 per 1,000 is shown as $10 per million. OpenAI's page says fine-tuning is closed to new users.
Worked example with the defaults: 2,000,000 tokens x 3 epochs = 6,000,000 training tokens. On Together AI, Qwen3.5 27B is 6 x $1.05 = $6.30, and Llama 3.1 8B is 6 x $0.34 = $2.04, which the $4 minimum lifts to $4.00. On Vertex AI, Gemini 3.5 Flash is 6 x $10 = $60.
Sources (read November 11, 2026)
Not sure whether to fine-tune or retrieve from your documents? See RAG pipeline development.
Frequently Asked Questions
Related Tools
Want expert help implementing this?
Our automation specialists can help you achieve these results and more. Book a free consultation to discuss your specific needs. Or see the exact service this tool maps to: RAG Pipeline Development.
Book a Free Consultation