modelwho

A lightweight CLI tool that helps developers choose the right AI model for their needs.

There are hundreds of AI models across OpenAI, Anthropic, Google, DeepSeek, Mistral, Meta, and dozens more providers. modelwho scores and ranks them based on what you actually care about (cost, capabilities, context window, reasoning, tool use) so you don't have to dig through docs and pricing pages.

Install

From source (requires Go 1.21+)

go install github.com/nehasriva/modelwho@latest

Build from repo

git clone https://github.com/nehasriva/modelwho.git
cd modelwho
make build
./bin/modelwho

Quick Start

# Interactive wizard — answers 2 questions, gives you recommendations
modelwho

# Search with filters
modelwho search --reasoning -n 5

# Use-case presets
modelwho search --preset coding

# Compare models side by side
modelwho compare claude-opus-4.5 gpt-5.2 deepseek-reasoner

# Detailed model info
modelwho info gpt-5-pro

How It Works

modelwho fetches model data and lets you search, filter, score, and compare models.

The scoring engine is the core differentiator. Instead of just filtering, it uses weighted multi-dimensional scoring across 7 dimensions:

Dimension	What it measures
Cost	Lower cost = higher score (inverse ranking)
Context Window	How well it meets your context needs
Reasoning	Chain-of-thought / extended thinking support
Tool Use	Function calling / tool calling support
Multimodal	Coverage of needed input modalities
Recency	How recently the model was updated
Output Capacity	Maximum output token limit

Weights are tuned per use case (presets) or set by the interactive wizard.

Commands

modelwho (no args) — Interactive Wizard

Launches a guided experience. You pick between:

• Quick mode — 2 questions (use case + budget), uses smart defaults
• Full mode — 5 pages to customize capabilities, context, budget, and provider preferences

modelwho search — Flag-Based Filtering

# Models with reasoning (Big 7 providers by default)
modelwho search --reasoning

# Budget-friendly models with tool calling
modelwho search --tool-call --max-cost 2

# Open-weight models that support images
modelwho search --open-weights --input-modality image

# Combine with presets — include all providers
modelwho search --preset agents --all -n 10

# Free-text search
modelwho search --query "claude" -n 5

Available flags:

Flag	Description
-p, --provider	Filter by provider(s) — openai, anthropic, google, etc.
-f, --family	Filter by model family
--reasoning	Require reasoning capability
--tool-call	Require tool calling
--structured-output	Require structured JSON output
--attachment	Require attachment support
--open-weights	Only open-weights / self-hostable models
--max-cost	Max input cost per million tokens
--max-cost-output	Max output cost per million tokens
--min-context	Minimum context window size
--min-output	Minimum output token limit
--input-modality	Required input types — text, image, audio, video, pdf
--all	Include all providers (default: Big 7 only)
--preset	Apply scoring preset — coding, agents, chatbot, etc.
-n, --top	Number of results (default 10)
-q, --query	Free-text search on model name/ID
--json	Output as JSON

modelwho compare — Side-by-Side

Compare 2–4 models with winner highlighting (star marks the best in each row):

$ modelwho compare claude-opus-4.5 gpt-5.2 deepseek-reasoner

  Model Comparison

       Claude Opus 4.5                   GPT-5.2                  DeepSeek Reasoner
          anthropic                       openai                       deepseek
 ────────────────────────────  ────────────────────────────  ────────────────────────────
 Cost In    $5.00              Cost In    $1.75              Cost In    $0.28 ★
 Cost Out   $25.00             Cost Out   $14.00             Cost Out   $0.42 ★
 Context    200K               Context    400K ★             Context    128K
 Output     64K                Output     128K ★             Output     128K ★
 ────────────────────────────  ────────────────────────────  ────────────────────────────
 Reasoning  ✓                  Reasoning  ✓                  Reasoning  ✓
 Tool Call  ✓                  Tool Call  ✓                  Tool Call  ✓
 Structured ✗                  Structured ✓                  Structured ✗
 Attachments ✓                 Attachments ✓                 Attachments ✓
 Open Wts   ✗                  Open Wts   ✗                  Open Wts   ✗
 ────────────────────────────  ────────────────────────────  ────────────────────────────
 Input:  text, image, pdf      Input:  text, image           Input:  text
 Updated: 2025-11-01           Updated: 2025-12-11           Updated: 2025-09-29

modelwho info — Model Details

modelwho info gpt-5-pro
modelwho info claude-sonnet   # partial match works

Shows pricing (input/output/cache), token limits, capabilities, modalities, dates, and which hosting providers also offer the model.

modelwho list — Browse

modelwho list providers          # all providers with model counts
modelwho list models             # all models
modelwho list models -p google   # just Google's models

modelwho update — Refresh Data

modelwho update   # force-refresh from models.dev

Data is cached at ~/.modelwho/cache.json with a 24-hour TTL. If you're offline, stale cache is used automatically with a warning.

Presets

Presets are pre-tuned scoring profiles for common use cases:

Preset	Best for	Key weights
coding	AI pair programming	High reasoning, tool use, context
creative	Writing & content	Balanced cost and quality
data	Data analysis / extraction	Reasoning, structured output, large context
chatbot	Customer support	Low cost, fast, conversational
agents	Autonomous tool use	Max tool calling + reasoning
multimodal	Image/audio/video	Multimodal input support
budget	Cheapest that works	Cost is everything

modelwho search --preset coding -n 5

Data Source

Model data comes from models.dev, an open-source MIT-licensed database maintained by the community. It covers 1,500+ models across 70+ providers including all major ones (OpenAI, Anthropic, Google, DeepSeek, Meta, Mistral, Cohere, xAI) and hosting providers (Azure, AWS Bedrock, Together AI, etc.).

By default, only the Big 7 providers are shown (OpenAI, Anthropic, Google, DeepSeek, Meta, Mistral, xAI). Use --all to include all hosting providers.

Project Structure

modelwho/
├── main.go                    # Entry point
├── cmd/                       # CLI commands (cobra)
│   ├── root.go                # No args → wizard, with flags → search
│   ├── search.go              # Flag-based filtering + scoring
│   ├── compare.go             # Side-by-side comparison
│   ├── list.go                # List providers/models
│   ├── info.go                # Detailed model view
│   ├── preset.go              # Use-case presets
│   ├── update.go              # Force cache refresh
│   └── wizard.go              # Interactive wizard orchestration
├── internal/
│   ├── api/                   # HTTP client + file cache
│   ├── model/                 # Domain types, catalog, filtering
│   ├── score/                 # Recommendation engine + presets
│   ├── wizard/                # Interactive forms (charmbracelet/huh)
│   └── ui/                    # Terminal rendering (lipgloss)
└── testdata/                  # API response fixture

Built With

• cobra — CLI framework
• huh — Interactive forms
• lipgloss — Terminal styling and tables
• models.dev — AI model database

License

MIT