LLM Leaders

The problem

Choosing a coding LLM today is a tab-hopping exercise: arena.ai tells you who is good at web dev, OpenRouter tells you what each model costs — per provider, per million tokens, changing constantly. Making a sane quality-per-dollar decision means holding those numbers in your head at the same time.

LLM Leaders does the holding for you.

What it does

One command renders the models you care about — or the entire OpenRouter catalog — with rank, price, and benchmark columns in a single terminal table:

# your curated list, sorted by OpenRouter Code rank
llm-leaders
 
# cheapest-input ≤ $1.50/M and arena rank ≤ 50, across the full catalog
llm-leaders --all --max-input 1.5 --max-rank 50
 
# markdown table for a doc or PR
llm-leaders --markdown

The Model column carries a value-for-money heat scale — Elo odds per dollar of blended price — so the quality-per-dollar story is visible, not implied. Free models with a known Elo render pure green; they are unbeatable per dollar.

How it stays honest

• Live prices from the OpenRouter catalog, refined to the cheapest provider from the endpoints API — the same "lowest across providers" price the website shows.
• Ranks from the arena.ai WebDev leaderboard, plus OpenRouter benchmark columns (27 categories) you can add with --bench.
• Coverage counters in the header (Code Rank/122) — a #1 in a sparse category is not a #1 of 122, and the table says so.
• Scales stay meaningful under filters: heat scales are computed over the rows actually displayed.
• Caching with purpose: prices 5 min (catalog) / 1 h (endpoints), arena and benchmarks 5 h — --refresh busts caches selectively.

Install

Arch users:

yay -S llm-leaders-bin

Everyone else builds from source with cargo build --release. Releases are published automatically from tagged commits — see the repository for details.