Model Transparency
Last reviewed: 2026-04-20 · v2.2.1-D
Every model ARIA routes to is declared here with context window, training cutoff, intended use cases, known limitations, and data-handling boundaries. Tenants can pin or forbid models per policy.
Claude Opus 4.7
frontieranthropic · ctx 1,000,000 · cutoff 2026-01
Use cases: complex reasoning, long-context analysis, governance decisions
Limitations: higher latency; higher cost per token
Data boundaries: Zero-retention API; no training on ARIA tenant data.
Claude Sonnet 4.5
balancedanthropic · ctx 200,000 · cutoff 2025-10
Use cases: default routing, content generation, customer copilots
Limitations: smaller context than Opus
Data boundaries: Zero-retention API; no training on ARIA tenant data.
GPT-4o
balancedopenai · ctx 128,000 · cutoff 2024-10
Use cases: multimodal intake, vision grounding
Limitations: older training cutoff; weaker long-context recall
Data boundaries: API tier with no training opt-in; standard DPA.
o1-mini
reasoningopenai · ctx 128,000 · cutoff 2024-10
Use cases: chain-of-thought reasoning, structured planning
Limitations: no tool use; higher latency
Data boundaries: API tier with no training opt-in.
BGE-small (embeddings)
locallocal · ctx 512 · cutoff 2024-06
Use cases: BYOK embeddings, on-droplet retrieval
Limitations: English-primary; multilingual quality lower
Data boundaries: Runs on ARIA droplet; data never leaves the tenancy.