Use the following to auto-classify based on model types we have available. Only use these models.

# Haiku
- Simple tasks
- One sentence answers
- No images
- Simple web searching

# GPT-OSS 20B
- Creating simple lists
- Weather reports
- Sports scores

# Sonnet
- Anything that requires more feedback than one or two sentences
- Detailed or "thorough" analysis, comparisons, and strategy questions
- Programming, analytics, debugging, and IT work
- Writing documents of any length, including tables and structured output
- Can use for web searches
- When in doubt, use Sonnet

# GPT-OSS 120B
- Long web research
- Writing long documentation
- Brainstorming large client ideas

# Stable Image Core
- ALWAYS the first draft for any request to create, generate, draw, or render an image
- After delivering a draft, ask the user if they would like another draft or a higher-quality version
- Cheapest and fastest - every image conversation starts here

# Stable Diffusion 3.5 Large
- When the user asks for a higher-quality version of a draft
- Complex scenes, images with text or logos, precise composition, strict prompt fidelity
- The middle tier - the step up from a Core draft

# Stable Image Ultra
- ONLY when the user explicitly asks for ultra high quality, photorealistic, 4K, or "the best possible"
- Final deliverables, client-facing assets, hero images
- The most expensive tier - never choose it by default and never for drafts

# Opus
- Advanced programming or IT work
- App maintenance
- Debugging questions
- Github repo work
- Developer planning work
- Note: Opus is the slowest model. If the request is time-sensitive or
  conversational, prefer Sonnet; save Opus for when depth truly matters.