Mint Recursive Service Pricing¶
Mint Recursive service billing and solution pricing¶
Mint Recursive provides three services: Training API, Inference API, and LLM as a judge. Mint Recursive training is priced separately. The table below lists Training API service rates during training. Ordinary hosted-model inference unit prices remain in the dashboard's Models & Pricing; see model unit prices. This page distinguishes training service rates, ordinary inference model rates, and solution quotes.
| Service | What it provides | Usage and access |
|---|---|---|
| Training API | Run training jobs on Mint Recursive, including supervised fine-tuning (SFT) and reinforcement learning (RL). | SFT, RL, or launch SFT through Mint Recursive Training. |
| Inference API | Generate model outputs through an API, including hosted model calls and sampling or inference with your own trained checkpoints. | Hosted model integrations and Using a trained model. |
| LLM as a judge | Use an LLM to evaluate outputs against a scoring rubric for evaluation, quality checks, and organizing feedback data. | Provide the outputs to evaluate, the scoring rubric, and the intended use of the results when discussing a service plan. |
Training service rates¶
Unit: USD per million tokens. This follows the workbook's unit label and dollar cell formatting; no currency conversion has been applied.
| Model | Algorithm | Prefill | Cached Prefill | Decode | Train |
|---|---|---|---|---|---|
| Qwen3.8 | SFT | $1.860 | $0.372 | $5.595 | $4.103 |
| GLM5.2 | SFT | $5.000 | $1.000 | $14.000 | $18.500 |
| GLM5.2 | RL | $5.000 | $1.000 | $14.000 | $24.050 |
| GLM5.3 | SFT | $5.000 | $1.000 | $14.000 | $18.500 |
| GLM5.3 | RL | $5.000 | $1.000 | $14.000 | $24.050 |
Dedicated inference after training (Dedicate)¶
All five Dedicate GPU entries in the source read TALK TO US. Contact sales for a quote; ordinary inference or training token rates do not determine this quote.
Solution inquiries¶
For dedicated inference, LLM as a judge, and other solutions, contact sales with the services you need, expected usage, and deployment requirements. Include a scoring rubric for judge workloads. Fees are subject to the agreed quote or order. For ordinary model API consumption, see Usage.