models

A fast CLI and TUI for browsing AI models, benchmarks, and coding agents.

• Models Tab: Browse 2000+ models across 85+ providers from models.dev, categorized by type (Origin, Cloud, Inference, Gateway, Dev Tool)
• Agents Tab: Track AI coding assistants (Claude Code, Aider, Cursor, etc.) with version detection and GitHub integration
• Benchmarks Tab: Compare model performance across 15+ benchmarks from Artificial Analysis, with creator filtering by source, region, and type

models-demo.mp4

What's New (v0.8.8)

Improved Open Weights Matching

• 91% match rate — three-stage matching pipeline using Jaro-Winkler similarity (strsim) to determine open/closed status per model
• Global fallback — when creator-scoped matching fails, searches all models.dev providers for the best slug match
• Known creator overrides — hardcoded open/closed status for 12 well-known creators absent from models.dev (IBM, AI2, TII, etc.)
• No more "Mixed" labels — removed CreatorOpenness fallback; unmatched models show an em dash instead of misleading labels

v0.8.7: Benchmark Data Freshness

• jsDelivr cache purging — GitHub Action now purges CDN cache after committing new data for faster propagation
• No disk cache — benchmark data fetched fresh from CDN on every launch for simplicity

v0.8.6: Cost Sorting & Open Weights

• Price sort columns — sort benchmarks by input, output, or blended price per million tokens via [s] cycle
• Per-model source detection — runtime matching of AA entries against models.dev data
• Source filter — [4] cycles through All / Open / Closed
• Region and type grouping — [5] and [6] toggle grouped layout with colored section headers

v0.8.5: Release Profile

• Optimized release binary — strip, LTO, single codegen unit, panic=abort (~6MB, down from ~11MB)

v0.8.0–0.8.4: Benchmarks Tab

• Dedicated Benchmarks tab — browse ~400 model entries from Artificial Analysis with quality, speed, and pricing data
• Creator sidebar with 40+ creators, classified by region and type with grouping toggles
• Quick-sort keys — [1] Intelligence, [2] Date, [3] Speed — press again to flip direction
• Dynamic columns, detail panel, TTFAT, AIME benchmarks and more

Other

• Provider categories — filter and group providers by type (Origin, Cloud, Inference, Gateway, Dev Tool)
• OpenClaw agent added to the agents catalog
• Responsive layouts — models tab detail panel scales with terminal height

Features

Models Tab

• CLI commands for scripting and quick lookups
• Interactive TUI for browsing and comparing models
• Provider categories — filter and group providers by type (Origin, Cloud, Inference, Gateway, Dev Tool)
• Cross-provider search to compare the same model across different providers
• Copy to clipboard with a single keypress
• JSON output for scripting and automation

Agents Tab

• Curated catalog of 12+ AI coding assistants
• Version detection — automatically detects installed agents
• GitHub integration — stars, releases, changelogs, update availability
• Persistent cache — instant startup with ETag-based conditional fetching
• Customizable tracking — choose which agents to monitor

Benchmarks Tab

• ~400 benchmark entries from Artificial Analysis with quality, speed, and pricing scores
• Auto-updating — data fetched fresh from CDN on every launch; GitHub Action refreshes source data every 6 hours
• Creator sidebar with 40+ creators — group by region or type with colored section headers
• Per-model open weights detection — runtime matching against models.dev, with source filter toggle
• Quick-sort keys — instantly sort by Intelligence, Date, or Speed
• Dynamic columns — list columns adapt to show the most relevant benchmarks for the active sort
• Detail panel — full benchmark breakdown with indexes, scores, performance, and pricing

Installation

Cargo (from crates.io)

cargo install modelsdev

Homebrew (macOS/Linux)

brew install arimxyer/tap/models

Scoop (Windows)

scoop bucket add arimxyer https://github.com/arimxyer/scoop-bucket
scoop install models

Pre-built binaries

Download the latest release for your platform from GitHub Releases.

Build from source

git clone https://github.com/arimxyer/models
cd models
cargo build --release
./target/release/models

Usage

TUI (Interactive Browser)

Just run models with no arguments to launch the interactive browser:

models

TUI Keybindings

Global

Key	Action
] / [	Switch tabs (Models / Agents / Benchmarks)
?	Show context-aware help
q	Quit

Navigation

Key	Action
j / ↓	Move down
k / ↑	Move up
g	Jump to first item
G	Jump to last item
Ctrl+d / PageDown	Page down
Ctrl+u / PageUp	Page up
Tab / Shift+Tab	Switch panels
← / →	Switch panels

Search

Key	Action
/	Enter search mode
Enter / Esc	Exit search mode
Esc	Clear search (in normal mode)

Models Tab

Filters & Sort

Key	Action
s	Cycle sort (name → date → cost → context)
1	Toggle reasoning filter
2	Toggle tools filter
3	Toggle open weights filter
4	Cycle provider category filter (All → Origin → Cloud → Inference → Gateway → Tool)
5	Toggle category grouping

Copy & Open

Key	Action
c	Copy provider/model-id
C	Copy model-id only
o	Open provider docs in browser
D	Copy provider docs URL
A	Copy provider API URL

Agents Tab

Filters & Sort

Key	Action
s	Cycle sort (name → updated → stars → status)
1	Toggle installed filter
2	Toggle CLI tools filter
3	Toggle open source filter

Actions

Key	Action
a	Open tracked agents picker
o	Open docs in browser
r	Open GitHub repo
c	Copy agent name

Customizing Tracked Agents

By default, models tracks 4 popular agents: Claude Code, Codex, Gemini CLI, and OpenCode.

Press a in the Agents tab to open the picker and customize which agents you track. Your preferences are saved to ~/.config/models/config.toml.

You can also add custom agents not in the catalog:

# ~/.config/models/config.toml
[[agents.custom]]
name = "My Agent"
repo = "owner/repo"
binary = "my-agent"
version_command = ["--version"]

See Custom Agents for the full reference.

Benchmarks Tab

Quick Sort (press again to toggle direction)

Key	Action
1	Sort by Intelligence index
2	Sort by Release date
3	Sort by Speed (tok/s)

Filters & Grouping

Key	Action
4	Cycle source filter (All / Open / Closed)
5	Toggle region grouping
6	Toggle type grouping

Sort (full cycle)

Key	Action
s	Cycle through all 20 sort columns
S	Toggle sort direction (asc/desc)

Actions

Key	Action
c	Copy benchmark name
o	Open Artificial Analysis page

CLI Commands

List providers

models list providers

List models

# All models
models list models

# Models from a specific provider
models list models anthropic

Show model details

models show claude-opus-4-5-20251101

Claude Opus 4.5
===============

ID:          claude-opus-4-5-20251101
Provider:    Anthropic (anthropic)
Family:      claude-opus

Limits
------
Context:     200k tokens
Max Output:  64k tokens

Pricing (per million tokens)
----------------------------
Input:       $5.00
Output:      $25.00
Cache Read:  $0.50
Cache Write: $6.25

Capabilities
------------
Reasoning:   Yes
Tool Use:    Yes
Attachments: Yes
Modalities:  text, image, pdf -> text

Metadata
--------
Released:    2025-11-01
Updated:     2025-11-01
Knowledge:   2025-03-31
Open Weights: No

Search models

models search "gpt-4"
models search "claude opus"

JSON output

All commands support --json for scripting:

models list providers --json
models show claude-opus-4-5 --json
models search "llama" --json

Data Sources

Lots of gratitude and couldn't have made this application without these workhorses doing the legwork. Shout out to the sources!:

• Model data: Fetched from models.dev, an open-source database of AI models maintained by SST
• Benchmark data: Fetched from Artificial Analysis — quality indexes, benchmark scores, speed, and pricing for ~400 model entries
• Agent data: Curated catalog in data/agents.json — contributions welcome!
• GitHub data: Fetched from GitHub API (stars, releases, changelogs)

License

MIT