Home / Glossary / LLM
What is Large Language Model?
Abbreviation: LLM
A large language model (LLM) is a neural network trained on vast text corpora to predict the next token, producing fluent natural-language output. osFoundry routes requests across 100+ LLMs from any provider — local, cloud, or self-hosted.
Detail
LLMs are transformer-based neural networks ranging from ~1B to over 1T parameters. They power chat, code generation, summarisation, translation, and most agentic AI today. Quality varies by training data, architecture, and post-training (RLHF, DPO).
LLMs come in two distribution modes: closed-source (accessed only via API — e.g. GPT-4, Claude) and open-weight (downloadable for self-hosting — e.g. Llama, Mistral, Qwen). osFoundry’s catalog indexes 76,000 open-weight models plus 364 hosted API models.
How osFoundry approaches Large Language Model
osFoundry treats every LLM as an interchangeable backend. Bring your own API key (BYOK) for hosted models; install open-weight models for local inference; deploy dedicated GPU endpoints for reserved capacity. Maestro routes per request based on rules you define in osStudio.
FAQ
How is osFoundry related to LLMs?
osFoundry is the orchestration layer above LLMs. It doesn’t make LLMs — it lets you route requests across local, cloud, and self-hosted LLMs from one workspace.
Which LLM is best?
No single LLM is best at everything. osFoundry lets you A/B different LLMs per workload and route accordingly.
Can I use multiple LLMs in one chat?
Yes. Maestro can switch models mid-conversation based on osStudio routing rules.
Related terms
Related features