Novita AI announced its language-model inference engine and published results for the hosted Llama 2 Chat 13B and 70B models. The service gave teams a hosted route to those weights without setting up their own language-model server.
Novita AI
Also known as: Novita · Novita.ai · Novita AI Cloud
Plain English
Novita AI offers hosted language and media models, dedicated deployments for custom language models, and rentable GPUs. It overlaps with fal.ai on media APIs and also serves teams that need a managed language-model endpoint.
Technical Overview
Shared chat endpoints expose an OpenAI-compatible API, while media jobs can use task-specific endpoints and return an ID for polling or webhook delivery. Dedicated language-model deployments can load custom weights and LoRA adapters. Teams that need control over the entire serving stack can instead run it on GPU instances.
Timeline (4)
2024
Release2025
EcosystemHugging Face added Novita to its Inference Providers. Users of the Hub's inference client could select Novita for a supported model through the client they already used for other providers.
- Ecosystem
Novita documented deployment of custom base models through its new LLM Dedicated Endpoint service. A team could deploy its own compatible model from Hugging Face and attach multiple LoRA adapters to one private endpoint, reusing the base model across variants.
2026
EcosystemNovita reported on its infrastructure work with SGLang, including GPU support for research and benchmarking and collaboration on large-scale expert parallelism. The collaboration gave the serving project infrastructure for testing how mixture-of-experts models distribute work across GPUs.
See also (8)
Open full graphNovita AI