[Feature Request] Model tiering — different models per task complexity within a single agent

Problem

Currently, the model provider (Anthropic or Azure OpenAI) is set at the agent level — all tasks on a given agent use the same model tier. There is no way to route simple tasks (health checks, compliance scans) to a cheaper model while reserving the premium model (Claude Opus) for complex incident investigations and RCA.

Current Behavior

• Model provider is configured in Settings → Basics at the agent level
• The agent auto-selects the best model within the chosen provider, but cannot switch providers per task
• To use different model tiers, teams must deploy separate agent instances — one per provider — which doubles always-on costs (4 AAU/hr per agent)

Proposed Solution

Allow per-task or per-thread model routing within a single agent instance:

1. Task-level model selection — e.g., scheduled health checks use GPT (cheaper), incident investigations use Claude Opus (deeper)
2. Complexity-based auto-routing — agent evaluates task complexity and selects the appropriate model tier automatically
3. Custom agent-level model override — allow sub-agents / custom agents to specify their own model provider, different from the parent agent

Use Case

Organizations running multiple SRE Agents across many products face a cost vs. depth trade-off. Each agent handles both:

• High-volume, low-complexity tasks: daily health checks, compliance scans, cost collection (~80% of tasks)
• Low-volume, high-complexity tasks: incident RCA, exception analysis, code-aware debugging (~20% of tasks)

Without model tiering, teams either:

• Pay premium rates (Claude Opus) for every task including simple ones, OR
• Deploy 2x agents per product (one GPT, one Claude) — doubling always-on baseline costs

Model tiering would let teams optimize cost without sacrificing investigation depth.

Impact

• High — direct cost optimization for multi-agent fleets
• Estimated 30-50% active flow cost reduction for organizations with mixed task complexity
• Aligns with the existing AAU pricing model which already differentiates rates by model

Environment

• Azure SRE Agent (GA)

No files, tests, or entry points are named. Start by tracing the agent-level provider configuration described under Settings → Basics and the task/thread execution path; define how provider selection, complexity routing, and custom-agent overrides should be represented and verified across the listed use cases.