model-recommender

name	model-recommender
description	Recommend the right model tier for each LLM task type
user-invocable	false

Model Recommender Skill

Task-to-Tier Mapping

Tier 1: Small / Fast (Haiku, GPT-4o-mini, Gemini Flash)

• Classification (spam/not-spam, category assignment)
• Entity extraction (names, dates, emails)
• Sentiment analysis
• Routing decisions
• Simple formatting (JSON restructuring, template filling)
• Yes/no judgments, slot filling

Tier 2: Mid-Range (Sonnet, GPT-4o, Gemini Pro)

• Summarization (single document)
• Translation
• Simple code generation (known patterns)
• Structured data extraction from complex documents
• Q&A with provided context
• Email/message drafting

Tier 3: Frontier (Opus, GPT-4/o1, Gemini Ultra)

• Complex multi-step reasoning
• Novel problem solving
• Nuanced creative writing
• Multi-document synthesis
• Architectural planning
• Novel/complex code generation

Decision Rules

1. When in doubt, start small — default to Tier 1.
2. Classification is ALWAYS Tier 1.
3. Extraction is almost always Tier 1.
4. Summarization is Tier 2 (unless multi-document → Tier 3).
5. Code generation: boilerplate = Tier 1, standard = Tier 2, novel = Tier 3.
6. If a task could be done with regex/rules → no LLM at all.

Red Flags

• Frontier model for classification → should be Tier 1
• Frontier model for formatting/extraction → should be Tier 1
• Frontier model for simple summarization → should be Tier 2
• Any LLM call for a deterministic task → should be code, not LLM