Agent Model Matching

The Core Insight

Think of AI models as developers on a team. Each has a different brain, different personality, different strengths.

A model isn’t just “smarter” or “dumber.” It thinks differently.Give the same instruction to Claude and GPT, and they’ll interpret it in fundamentally different ways.

Oh My OpenCode assigns each agent a model that matches its working style—like building a team where each person is in the role that fits their personality.

How Claude and GPT Think Differently

Claude
GPT (5.2+)

Mechanics-Driven

Best with:

Detailed checklists
Step-by-step procedures
Explicit templates
Nested workflows

Philosophy: More rules = more compliance. You can write a 1,100-line prompt with nested workflows and Claude will follow every step.Example agent: Prometheus’s Claude prompt is ~1,100 lines across 7 files.

Agents that support both families (Prometheus, Atlas) auto-detect your model at runtime and switch prompts.

Agent Profiles

Communicators → Claude / Kimi / GLM

These agents have Claude-optimized prompts—long, detailed, mechanics-driven. They need models that reliably follow complex, multi-layered instructions.

Sisyphus

Role: Main orchestrator
Model: Claude Opus 4.6Why Claude:

Follows complex multi-step instructions (prompt is ~1,100 lines)
Maintains conversation flow across many tool calls
Understands nuanced delegation patterns
Produces well-structured output

Fallback chain:

Claude Opus → Kimi K2.5 → GLM 5

Notes:

No GPT prompt exists — Sisyphus is Claude-family only
Using Sisyphus with GPT would be like taking your best project manager and sticking them in a room alone to debug a race condition

Metis

Role: Plan gap analyzer
Model: Claude Opus 4.6Why Claude:

Excellent at finding ambiguities
Detects AI-slop patterns
Identifies hidden intentions

Fallback chain:

Claude Opus → Kimi K2.5 → GPT-5.2 → Gemini 3 Pro

Notes: Claude preferred, GPT acceptable fallback.

Dual-Prompt Agents → Claude preferred, GPT supported

These agents ship separate prompts for Claude and GPT families. They auto-detect your model and switch at runtime.

Prometheus

Role: Strategic planner
Model: Claude Opus 4.6 (with extended thinking)Why dual-prompt:

Claude: detailed mechanics for interview process
GPT: compact principle-driven planning

Fallback chain:

Claude Opus → GPT-5.2 → Kimi K2.5 → Gemini 3 Pro

Auto-detection:

if (isGptModel(modelID)) {
  return GPT_PROMETHEUS_PROMPT  // ~121 lines
} else {
  return CLAUDE_PROMETHEUS_PROMPT  // ~1,100 lines
}

Atlas

Role: Todo orchestrator
Model: Kimi K2.5Why Kimi: Claude-like behavior at lower cost. Sweet spot for orchestration.Fallback chain:

Kimi K2.5 → Claude Sonnet → GPT-5.2

Auto-detection: Switches between Claude-optimized and GPT-optimized prompts.

Deep Specialists → GPT

These agents are built for GPT’s principle-driven style. Their prompts assume autonomous, goal-oriented execution.

Hephaestus

Role: Autonomous deep worker
Model: GPT-5.3 Codex (medium reasoning)Why GPT:

Deep autonomous exploration
Multi-file reasoning
Principle-driven execution (goal, not recipe)
Works independently for extended periods

Fallback chain:

GPT-5.3 Codex only (no fallback)

Notes:

Do not override to Claude. Hephaestus is built for Codex’s autonomous style.
Named after Greek god of forge and craftsmanship.
Inspired by AmpCode’s deep mode.

Oracle

Role: Architecture consultant
Model: GPT-5.2Why GPT:

High-IQ strategic reasoning
Deep logical analysis
Read-only consultation

Fallback chain:

GPT-5.2 → Gemini 3 Pro → Claude Opus

Momus

Role: Ruthless plan reviewer
Model: GPT-5.2Why GPT:

Verification and critique
Different perspective from Claude
Rigorous logic

Fallback chain:

GPT-5.2 → Claude Opus → Gemini 3 Pro

Utility Runners → Speed over Intelligence

These agents do grep, search, and retrieval. They intentionally use the fastest, cheapest models available.

Don’t “upgrade” utility agents to Opus. That’s hiring a senior engineer to file paperwork.

Explore

Role: Fast codebase grep
Model: Grok Code Fast 1Why fast/cheap:

Speed is everything
Fire 10 in parallel
Simple pattern matching

Fallback chain:

Grok Code Fast → MiniMax → Haiku → GPT-5-Nano

Librarian

Role: Docs/code search
Model: Gemini FlashWhy fast/cheap:

Doc retrieval doesn’t need deep reasoning
High volume of searches

Fallback chain:

Gemini Flash → MiniMax → GLM

Multimodal Looker

Role: Vision/screenshots
Model: Kimi K2.5Why Kimi:

Excels at multimodal understanding
Good image analysis

Fallback chain:

Kimi K2.5 → Gemini Flash → GPT-5.2 → GLM-4.6v

Model Families

Claude Family

Characteristics:

Communicative
Instruction-following
Structured output

Model	Strengths	Use For
Claude Opus 4.6	Best overall. Highest compliance with complex prompts.	Sisyphus, Prometheus
Claude Sonnet 4.6	Faster, cheaper. Good balance.	Everyday tasks
Claude Haiku 4.5	Fast and cheap.	Quick tasks, utility work
Kimi K2.5	Claude-like behavior at lower cost.	Atlas, all-rounder
GLM 5	Claude-like, solid for orchestration.	Cost-effective orchestration

GPT Family

Characteristics:

Principle-driven
Explicit reasoning
Deep technical capability

Model	Strengths	Use For
GPT-5.3 Codex	Deep coding powerhouse. Autonomous exploration.	Hephaestus (required)
GPT-5.2	High intelligence, strategic reasoning.	Oracle, Momus
GPT-5-Nano	Ultra-cheap, fast.	Simple utility tasks

Other Models

Model	Strengths	Use For
Gemini 3 Pro	Visual/frontend tasks. Different reasoning style.	`visual-engineering`, `artistry`
Gemini 3 Flash	Fast. Doc search and light tasks.	Librarian
Grok Code Fast 1	Blazing fast code grep.	Explore
MiniMax M2.5	Fast and smart. Utility tasks.	Fallback for search/retrieval

Free-Tier Fallbacks

You may see model names like kimi-k2.5-free, minimax-m2.5-free, or big-pickle (GLM 4.6) in logs. These are free-tier versions served through OpenCode Zen provider.

You don’t need to configure free-tier fallbacks. The system includes them automatically so it degrades gracefully when you don’t have every paid subscription.

Task Categories

When agents delegate work, they don’t pick a model name—they pick a category. The category maps to the right model automatically.

Category	When Used	Fallback Chain
`visual-engineering`	Frontend, UI, CSS, design	Gemini 3 Pro → GLM 5 → Claude Opus
`ultrabrain`	Maximum reasoning needed	GPT-5.3 Codex → Gemini 3 Pro → Claude Opus
`deep`	Deep coding, complex logic	GPT-5.3 Codex → Claude Opus → Gemini 3 Pro
`artistry`	Creative, novel approaches	Gemini 3 Pro → Claude Opus → GPT-5.2
`quick`	Simple, fast tasks	Claude Haiku → Gemini Flash → GPT-5-Nano
`unspecified-high`	General complex work	Claude Opus → GPT-5.2 → Gemini 3 Pro
`unspecified-low`	General standard work	Claude Sonnet → GPT-5.3 Codex → Gemini Flash
`writing`	Text, docs, prose	Gemini Flash → Claude Sonnet

See the Orchestration Guide for how agents dispatch tasks to categories.

Customization

Example Configuration

{
  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",

  "agents": {
    // Main orchestrator: Claude Opus or Kimi K2.5 work best
    "sisyphus": {
      "model": "kimi-for-coding/k2p5",
      "ultrawork": { 
        "model": "anthropic/claude-opus-4-6", 
        "variant": "max" 
      }
    },

    // Research agents: cheaper models are fine
    "librarian": { "model": "zai-coding-plan/glm-4.7" },
    "explore":   { "model": "github-copilot/grok-code-fast-1" },

    // Architecture consultation: GPT or Claude Opus
    "oracle": { 
      "model": "openai/gpt-5.2", 
      "variant": "high" 
    },

    // Prometheus inherits sisyphus model; just add prompt guidance
    "prometheus": { 
      "prompt_append": "Leverage deep & quick agents heavily, always in parallel." 
    }
  },

  "categories": {
    "quick": { "model": "opencode/gpt-5-nano" },
    "unspecified-low": { "model": "kimi-for-coding/k2p5" },
    "unspecified-high": { 
      "model": "anthropic/claude-sonnet-4-6", 
      "variant": "max" 
    },
    "visual-engineering": { 
      "model": "google/gemini-3-pro", 
      "variant": "high" 
    },
    "writing": { "model": "kimi-for-coding/k2p5" }
  },

  // Limit expensive providers; let cheap ones run freely
  "background_task": {
    "providerConcurrency": { 
      "anthropic": 3, 
      "openai": 3, 
      "opencode": 10, 
      "zai-coding-plan": 10 
    },
    "modelConcurrency": { 
      "anthropic/claude-opus-4-6": 2, 
      "opencode/gpt-5-nano": 20 
    }
  }
}

Safe vs Dangerous Overrides

✅ Safe
❌ Dangerous

Same personality type:

{
  "agents": {
    // Sisyphus: Opus → Sonnet, Kimi K2.5, GLM 5
    // All communicative models
    "sisyphus": { "model": "kimi-for-coding/k2p5" },
    
    // Prometheus: Opus → GPT-5.2
    // Auto-switches to GPT prompt
    "prometheus": { "model": "openai/gpt-5.2" },
    
    // Atlas: Kimi K2.5 → Sonnet, GPT-5.2
    // Auto-switches to GPT prompt
    "atlas": { "model": "anthropic/claude-sonnet-4-6" }
  }
}

Personality mismatch:

{
  "agents": {
    // ❌ Sisyphus → GPT
    // No GPT prompt exists. Will degrade significantly.
    "sisyphus": { "model": "openai/gpt-5.2" },
    
    // ❌ Hephaestus → Claude
    // Built for Codex's autonomous style. Claude can't replicate.
    "hephaestus": { "model": "anthropic/claude-opus-4-6" },
    
    // ❌ Explore → Opus
    // Massive cost waste. Explore needs speed, not intelligence.
    "explore": { "model": "anthropic/claude-opus-4-6" },
    
    // ❌ Librarian → Opus
    // Same. Doc search doesn't need Opus-level reasoning.
    "librarian": { "model": "anthropic/claude-opus-4-6" }
  }
}

Model Resolution

Each agent has a fallback chain. The system tries models in priority order until it finds one available through your connected providers.

Agent Request → User Override (if configured) → Fallback Chain → System Default

You don’t need to configure providers per model—just authenticate (opencode auth login) and the system figures out which models are available and where.

Checking Available Models

# List all models you have access to
opencode models

# Run diagnostics
bunx oh-my-opencode doctor --verbose

Common Questions

Can I use GPT for Sisyphus?

No. Sisyphus has no GPT prompt. It’s designed for Claude’s mechanics-driven instruction-following. Using GPT will significantly degrade performance.If you want GPT-style reasoning, use Hephaestus instead.

Why does Hephaestus require GPT?

Hephaestus is built around GPT-5.3 Codex’s autonomous exploration style. Its prompt assumes:

Goal-oriented execution
Minimal hand-holding
Deep independent reasoning

Claude’s mechanics-driven style doesn’t match this working pattern.

Which models can I use for Prometheus?

Prometheus auto-detects Claude vs GPT and switches prompts accordingly. Safe options:

Claude: Opus, Sonnet, Kimi K2.5, GLM 5
GPT: GPT-5.2, GPT-5.3 Codex
Gemini: Gemini 3 Pro (uses Claude-style prompt)

How do I make all agents cheaper?

Target utility agents first:

{
  "agents": {
    "explore": { "model": "opencode/gpt-5-nano" },
    "librarian": { "model": "opencode/gpt-5-nano" }
  },
  "categories": {
    "quick": { "model": "opencode/gpt-5-nano" }
  }
}

Keep Sisyphus/Prometheus on quality models—they’re worth it.

Custom Categories

Define domain-specific model presets

Background Agents

Concurrency limits per model/provider

Prometheus Planning

Why Prometheus uses Claude Opus with extended thinking

Configuration

Full agent configuration reference

Get Started

Core Concepts

Guides

Configuration

The Core Insight

How Claude and GPT Think Differently

Mechanics-Driven

Principle-Driven

Agent Profiles

Communicators → Claude / Kimi / GLM

Dual-Prompt Agents → Claude preferred, GPT supported

Deep Specialists → GPT

Utility Runners → Speed over Intelligence

Model Families

Claude Family

GPT Family

Other Models

Free-Tier Fallbacks

Task Categories

Customization

Example Configuration

Safe vs Dangerous Overrides

Model Resolution

Checking Available Models

Common Questions

Custom Categories

Background Agents

Prometheus Planning

Configuration

Build docs developers (and LLMs) love

Get Started

Core Concepts

Guides

Configuration

​The Core Insight

​How Claude and GPT Think Differently

​Mechanics-Driven

​Principle-Driven

​Agent Profiles

​Communicators → Claude / Kimi / GLM

​Dual-Prompt Agents → Claude preferred, GPT supported

​Deep Specialists → GPT

​Utility Runners → Speed over Intelligence

​Model Families

​Claude Family

​GPT Family

​Other Models

​Free-Tier Fallbacks

​Task Categories

​Customization

​Example Configuration

​Safe vs Dangerous Overrides

​Model Resolution

​Checking Available Models

​Common Questions

​Related

Custom Categories

Background Agents

Prometheus Planning

Configuration

Build docs developers (and LLMs) love

The Core Insight

How Claude and GPT Think Differently

Mechanics-Driven

Principle-Driven

Agent Profiles

Communicators → Claude / Kimi / GLM

Dual-Prompt Agents → Claude preferred, GPT supported

Deep Specialists → GPT

Utility Runners → Speed over Intelligence

Model Families

Claude Family

GPT Family

Other Models

Free-Tier Fallbacks

Task Categories

Customization

Example Configuration

Safe vs Dangerous Overrides

Model Resolution

Checking Available Models

Common Questions

Related