TritonGPT Large Language Models
TritonGPT provides access to multiple models through approved UC San Diego routes. The available catalog changes as models are evaluated, added, updated, or retired.
Use the Model Hub for the current catalog
The Model Hub is the authoritative source for models currently offered through the TritonAI gateway, including published capabilities, context windows, and rates.
Choose from the task outward
Do not select a model from a universal ranking. Start with the work and the information involved:
- Task: What must the model produce, and how will the result be checked?
- Sources: Does the work require current, approved knowledge or documents supplied by the user?
- Data: What classification and project-specific controls apply?
- Capability: Does the task need long context, multimodal input, structured output, tool use, speed, or another specific capability?
- Cost and scale: What do the current published rates and expected usage imply?
Approved hosting routes
The TritonAI gateway can provide approved enterprise cloud models and open models hosted on UC-controlled infrastructure. The appropriate route depends on the service, data, capability, and controls; neither route should be treated as automatically approved for every workload.
Keep review in the workflow
Model outputs can be incomplete, outdated, or wrong. Verify important claims against authoritative sources, keep consequential decisions with accountable people, and follow the access and data-handling conditions confirmed for the project.