Models
Serverless prices are set live by the sourcing engine — always at or below the reference platform's list price, never below our cost.
Frontier open-weights MoE model with strong agentic tool-calling and long-context reasoning.
Trillion-parameter MoE tuned for agentic workloads and coding.
Efficient MoE with hybrid reasoning modes and strong math/code performance.
Hybrid thinking-mode MoE, excellent multilingual and reasoning trade-off.
Natively multimodal MoE with 1M-token context window.
Workhorse dense instruct model; great quality-per-dollar for fine-tuning.
Code-specialized model with strong repo-level editing and FIM support.
Small, fast, cheap — ideal for classification and high-volume tasks.
State-of-the-art open image generation, $0.0014/step reference pricing.
Fast speech transcription, $0.0015/audio-minute reference pricing.
High-recall embedding model for RAG and semantic search.