There is no best model, and a single leaderboard mostly measures whichever benchmark the ranker weighted. Separating coding from maths from vision at least asks a question that can be answered, and putting pricing next to the benchmarks means the answer accounts for what it costs. The obvious caveat applies to anything in this category: model releases move faster than any curated comparison, so treat the numbers as a starting point and check the current pricing before committing. The recommendation tool takes a budget, which is the right input — most model choices are cost-constrained rather than capability-constrained.
A model selection advisor with four tools: detailed information on one model, top rankings by category, side-by-side comparison, and a recommendation from stated requirements.
- get_model_info returns pricing and benchmark results for a specific model rather than a description of it.
- list_top_models ranks by category — coding, math and vision among them — so the question is answered per task rather than in general.
- compare_models puts two to five side by side on pricing, benchmarks and capabilities.
- recommend_model takes a use case, a budget and requirements and returns a recommendation rather than a list.
Nothing to supply.
One command — claude mcp add llm-advisor -- npx -y llm-advisor-mcp