Intelligent model routing that cuts LLM token costs and latency
Open-source engine for faster local language model inference with bit-for-bit identical output
Stage
Not given
Not given
Founded
Not given
Not given
Based in
Not given
Not given
Pricing model
Paid
Open source
Pricing
Pay-as-you-go with 1% service fee. Cheap Tasks: $0.000043–$0.000194 per reference request. Standard Tasks: $0.000225–$0.000838 per reference request. Reasoning Tasks: $0.00066–$0.0023 per reference request. Frontier Tasks: $0.0023–$0.0354 per reference request. Agentic Tasks: $0.0071–$0.0354 per ref
Not given
Key features
Automatic task classification with Jev decision model
Multi-provider model access (OpenRouter, Kie.ai, Replicate, fal.ai)
Lowest-price eligible model selection
Support for text, image, video and audio modalities
Budget limits and provider filtering
Transparent billing and cost tracking
Three question types: Choice, Score and Noul
Pay-as-you-go pricing with no subscription
Speculative decoding for faster token generation
Prompt caching for context reuse
Real-time prompt processing while typing
Concurrent request batching
Bit-for-bit identical output verification
Multi-GPU and multi-backend support (CUDA, Vulkan)
Vision model support with image encoding
Mixture-of-experts model support
RAD-1 automatic draft token generation
Cross-platform support for Windows and Linux
What makes it different
Jev decision model automatically matches tasks to appropriate capability tiers instead of fixed frontier model routing
Reduces costs by up to 43.9% on illustrative workloads compared to fixed-model routing
Single API key for four different providers with unified billing
Works with multiple modalities and use cases like triage, classification and gating
Exact deterministic inference: all default optimizations preserve bit-for-bit identical output compared to plain decoding
4.5× faster performance on editing tasks on small models compared to llama.cpp
Verifiable reproducibility through receipt system that records all request parameters
Nine optimization techniques enabled by default with ability to disable each individually
No dependency on draft models or external selections for speculative decoding
Token Cut
Kinetic