Learning path · Models & Tokens · 11
Model Families
Lineages of models sharing architecture, tokenizer, and training recipe—GPT, Claude, Gemini, Llama, Mistral, and domain-specific variants.
Why it matters
- Capabilities, context limits, tool formats, and pricing differ by family.
- Migration is rarely drop-in: prompts and evals must be revalidated.
- Routing across families is a common enterprise pattern.
Key ideas
- Capability tiers
- Context and tool support
- Vendor roadmaps
Model families differ in reasoning strength, instruction following, multimodal support, structured output fidelity, and safety behaviour. A prompt tuned for one family may underperform on another. Build abstraction layers for prompts, tool schemas, and eval suites so you can swap models without rewriting orchestration. Keep a small golden task set per family to detect regressions when providers ship silent updates. Publish an internal compatibility matrix listing supported tool formats, max context, and deprecation dates so squads plan migrations early. Ship only after eval gates pass on representative production failures.
Updated 2026-08-09 · Full learning path