Billing
Overview of Billing Components
Fine-tuning jobs are billed based on the number of tokens processed and the model size tier.
1. Token-Based Pricing
Charged per 1M tokens, calculated as the total tokens processed during training (sum across all epochs), plus tokens used in the optional validation dataset.
2. Model Size Tier
Pricing is grouped by model size. For example, models up to 10B parameters share the same token rate.
Model size tier | SKU | Price ($ per 1M Tokens) |
---|---|---|
Up to 10B parameters | ogc-ft-10b-tokens | 0.46 |
All fine-tuning uses LoRA, enabling faster training at lower compute cost.
Example Usage & Billing
Let’s say you’re fine-tuning a model with 8B parameters, which falls under OGC’s “up to 10B” pricing tier and billed at $0.00000046 per token (i.e., $0.46 per 1M tokens).
- Your training dataset contains 5 million tokens
- You run 3 epochs → So total processed = 5M × 3 = 15 million tokens
- You also include a validation set with 1 million tokens
Total tokens billed: 15M (training) + 1M (validation) = 16 million tokens
Billing Calculation:
16M tokens × $0.00000046 = $7.36
Your total cost for this fine-tuning job would be $7.36