One API for 72 leading AI models from 15 labs with unified billing and automatic failover
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
NZ
Not given
Pricing model
Paid
Open source
Pricing
Prepaid USD balance with per-token pricing. Text models from $0.627 input / $1.881 output per 1M tokens. Image generation from $0.011 per image. Video generation from $0.043 per second. Card top-ups from $5 with 5.5% fee ($0.80 minimum). Trial credit: $1 for 7 days.
Not given
Key features
OpenAI and Anthropic API format compatibility
Automatic failover routing to backup routes
Bring your own keys (BYOK) with 1M free monthly requests
Unified USD billing with per-token pricing
Image and video generation support
Team and project-level access controls
Usage tracking with per-model details and exports
Support for streaming with standard server-sent events
Automatic route verification at least weekly
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
All 72 models through one API with unified billing, no separate accounts needed
Leading models priced at or below official list prices
Automatic failover keeps requests on the same model even when a route fails
Free BYOK plan with 1M monthly requests using your own provider keys
Per-token billing with token counts and prices shown per request
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
BoostRail
Kinetic