Self-serve access with no base fee via pre-paid credits, billed per token, GPU second or training token.
Serverless inference per 1M tokens, split into input, cached input and output
Kimi K3 and Ember-1: USD 3.00 input, USD 15.00 output per 1M tokens
GLM 5.3: USD 1.40 input, USD 4.40 output per 1M tokens
DeepSeek V4.1 Flash and MiniMax M3: USD 0.30 input, USD 1.20 output per 1M tokens
OpenAI gpt-oss-120B and GLM 5.3 Flash: USD 0.15 input, USD 0.50 to 0.60 output per 1M tokens
NVIDIA Nemotron 3.5 Lightning 30B: USD 0.05 input, USD 0.20 output per 1M tokens
Other models by size: USD 0.10 (under 4B parameters), USD 0.20 (4B to 16B), USD 0.90 (over 16B) per 1M tokens
Priority mode, prioritized under high load, at a premium, Fast variants of selected models
Batch inference at 50% of serverless pricing
Embeddings from USD 0.008 per 1M input tokens
Fine-tuning (SFT and DPO) from USD 0.50 per 1M training tokens, fine-tuned models served at base model prices
Serverless Training API for LoRA, billed per token for prefill, sampling and training
On-demand GPUs billed per second: H100 and H200 USD 8.00, B200 USD 13.00, B300 USD 15.00, GB300 USD 20.00 per hour
Pre-paid credits with Auto Reload and a monthly spend limit
