llm-model-compare

Compare LLMs on OpenRouter by quality (intelligence / coding / agentic) and price — one table, one command.

Runs on Bun. No API key required (one anonymous request to the OpenRouter model catalogue); set a key to lift rate limits.

Usage

# one or more model ids, slugs, or globs
bun index.ts anthropic/claude-opus-5 openai/gpt-5.6-sol

# a JSON list
bun index.ts models.json

# comma-separated patterns
bun index.ts --models "openai/*,x-ai/grok-*"

# discover models by keyword
bun index.ts --search coding --top 10

# machine-readable output
bun index.ts --json --models "deepseek/*"

Output:

model                            score  intel  coding  agentic  context  in $/M  out $/M
-------------------------------  -----  -----  ------  -------  -------  ------  -------
anthropic/claude-opus-5           66.8     63      78       59       1M   $5.00   $25.00
openai/gpt-5.6-sol                65.4     61      77       58    1.05M   $2.00   $10.00
qwen/qwen3.8-2.4t-a95b            62.2     58      72       57    1.05M   $2.00    $6.00

score = mean of intel/coding/agentic (OpenRouter artificial_analysis, 0-100)

• score — mean of the three quality indices (0–100)
• intel / coding / agentic — OpenRouter's own artificial_analysis indices
• context — context window
• in $/M / out $/M — USD per 1M input/output tokens

The models JSON

Any of:

["deepseek/deepseek-v4-pro-0813", "qwen/qwen3.8-27b"]

{ "models": ["anthropic/claude-opus-5", "openai/*"] }

{ "enabledModels": ["claude-*", "gpt-4o"] }

Entries may be full openrouter/author/model ids, bare author/model slugs, or * / ? globs. Rolling ~… aliases (e.g. ~deepseek/deepseek-v4-flash-latest) are followed to their concrete model's scores.

Install as a binary

npm install -g llm-model-compare
llm-model-compare --models "openai/*"

Options

flag	description
--models "a/b,c/*"	comma-separated ids/globs
--search <word>	substring search across id/name/description
--top <n>	limit --search results (default 20)
--json	JSON output instead of a table
--key <sk-…>	OpenRouter API key (OPENROUTER_API_KEY also works)
-h, --help	help
-v, --version	version

How scores work

The score column is the arithmetic mean of three indices OpenRouter publishes under benchmarks.artificial_analysis: intelligence_index, coding_index, and agentic_index. They are OpenRouter's own ratings, not this tool's opinion — models without scores show -.

License

MIT