Skip to main content
Access to AI models is one of the cross-cutting concerns Governe handles once for the whole platform, so every agent draws from the same governed catalog. This page lists the models enabled by default on the Prisme.ai platform, grouped by provider. Use it as a reference when configuring agents, RAG pipelines, or routing rules.
The catalog below reflects the default platform configuration. Available models in your environment may differ depending on:
  • Provider credentials configured by your platform administrator.
  • Organization-level restrictions set in Model Governance.
  • Per-agent model restrictions.
To check what is actually allowed for your organization, go to Agents Controls > Models in AI Governance.

​
Transcription model

gpt-4o-transcribe, on Azure OpenAI in the EU Data Zone, is the platform default transcription model. It powers chat dictation and the LLM Gateway POST /v1/audio/transcriptions endpoint. If your organization restricts its allowed models, keep a transcription model in the list. Otherwise members dictate with the browser’s own speech service in desktop Chrome and Edge, which sends the audio to Google or Microsoft, unless the browser dictation fallback is turned off, and have no microphone button in other browsers. Turn off Allow browser dictation fallback in Agents Controls → Models if leaving the model out is meant to turn dictation off. See Transcription model for dictation.

​
Cost tiers

Each completion model is tagged with a cost tier used by routing strategies (e.g. Cost Optimized) and by the soft-downgrade policy:

​
Catalog

Switch between viewing the catalog grouped by provider or by region.
  • Model Governance: restrict allowed models, configure defaults, set quotas, define routing rules, and configure failover.
  • Capabilities: manage tools, guardrails, skills, and memory available to agents.
  • LLM Gateway API: programmatic access to all models through a single endpoint.