Choose the mind.
Keep the workspace.

20 models from 8 labs, grouped by the work they fit. Start with a recommendation, see how models pair across a session, or filter the full directory.

  • OpenAI
  • Anthropic
  • Google
  • DeepSeek
  • Meta
  • xAI
  • Mistral
  • Qwen

WHERE TO START / 5 KINDS OF WORK

Match the model
to the task.

  1. 01

    DAILY CODING

    Everyday coding

    Strong all-round models for multi-file edits, feature work, and most tasks you hand off.

    • Claude Sonnet 4.5
    • GPT-5
    • Qwen 3 Max
    • Mistral Large
    • GPT-4.1
    Filter →
  2. 02

    REASONING

    Hard problems

    Deliberate models for difficult bugs, design tradeoffs, and plans that need to hold up.

    • Claude Opus 4.5
    • o3
    • DeepSeek R1
    • Grok 4
    • o4-mini
    • Magistral
    Filter →
  3. 03

    VISION & CONTEXT

    Interfaces & large repos

    Models that read screenshots and long context, for visual checks and repository-wide work.

    • Gemini 3 Pro
    • Gemini 2.5 Pro
    • Qwen 3 VL
    Filter →
  4. 04

    FAST

    Quick iterations

    Responsive models for small edits, lint fixes, test triage, and log summaries.

    • Claude Haiku 4.5
    • Gemini 2.5 Flash
    • Grok 3 Mini
    Filter →
  5. 05

    OPEN WEIGHT

    Open ecosystems

    Open-weight models for teams that prefer open tooling and high-volume work.

    • Llama 4 Maverick
    • DeepSeek V3
    • Llama 4 Scout
    Filter →

MULTI-MODEL ROUTING / 3 PATTERNS

One workspace.
Multiple minds.

Because files, build caches, and booted simulators live on the cloud Mac rather than inside a chat window, different models can take turns on the same task without losing state.

01APPLE PLATFORMS

Native SwiftUI & Simulator QA

Use a reasoning model to untangle layout constraints, a flagship coder to patch Swift files and run XCTest, and a vision model to inspect the booted iPhone screen.

  1. PLANo3Trace detent sizing
  2. BUILDClaude Sonnet 4.5Patch & xcodebuild
  3. VERIFYGemini 3 ProSimulator visual QA
Inspect in the harness →
02WEB & SAFARI

WebKit Compositor & Cross-Browser E2E

Isolate Safari-only layer promotion bugs with auditable reasoning, refactor CSS and TypeScript, and verify zero layout shift across WebKit and Chromium.

  1. PLANDeepSeek R1Isolate GPU layer bug
  2. BUILDGPT-5Refactor sticky header
  3. VERIFYQwen 3 VL0px shift check
Inspect in the harness →
03SYSTEMS & CLI

Apple Silicon CLI & System Frameworks

Adapt to macOS Security.framework APIs, compile native arm64 Mach-O binaries, and validate Homebrew bottle formulas in a clean shell.

  1. PLANGrok 4Map Keychain Cgo API
  2. BUILDMistral LargeBuild arm64 binary
  3. VERIFYLlama 4 Maverickbrew test audit
Inspect in the harness →

FULL DIRECTORY / 20

Every model.

  • Claude Sonnet 4.5 · Anthropic
  • Claude Opus 4.5 · Anthropic
  • GPT-5 · OpenAI
  • o3 · OpenAI
  • Gemini 3 Pro · Google
  • DeepSeek R1 · DeepSeek
  • Qwen 3 Max · Qwen
  • Llama 4 Maverick · Meta
  • Grok 4 · xAI
  • Mistral Large · Mistral
  • Claude Haiku 4.5 · Anthropic
  • GPT-4.1 · OpenAI
  • o4-mini · OpenAI
  • Gemini 2.5 Pro · Google
  • Gemini 2.5 Flash · Google
  • DeepSeek V3 · DeepSeek
  • Llama 4 Scout · Meta
  • Grok 3 Mini · xAI
  • Magistral · Mistral
  • Qwen 3 VL · Qwen

Available models may change during the private beta. Model names are trademarks of their respective owners.

Models are only half the agent.

See how the JY Code harness grounds context, dispatches macOS tools, and verifies every edit.

Explore the harness