Overview
Compare the Claude Code, Codex, and Cursor harnesses that Conductor can run
Conductor can run Claude Code, Codex, and Cursor sessions in isolated workspaces. A harness is the provider agent that writes code; Conductor is the workspace layer around that agent.
Conductor is free to use today. Conductor does not bill, meter, or resell model usage. Claude Code, Codex, and Cursor usage is billed through the provider account, subscription, credits, usage pool, or API token you configure.
Provider pricing and model access can change. Use this page to understand how Conductor connects each harness, then check the linked provider docs for current usage limits and pricing.
Harness comparison
Choose the provider agent by workflow, setup path, and billing owner. Conductor stays the workspace layer.
Models at a glance
Model access changes upstream. Use this as a local reference, then check the provider docs for current availability.
Claude Code7 models, starting with Opus 4.8
| Model | Guidance |
|---|---|
| Opus 4.8 | Newest Opus option listed in Anthropic's Claude Code model configuration. Availability depends on your Claude Code account, subscription, API key, and provider setup. |
| Opus 4.7 | Versioned Opus option for deeper reasoning and harder coding tasks. Use the /model or /status commands in Claude Code to confirm whether your account exposes it. |
| Sonnet 4.6 | Current Sonnet option listed in Anthropic's Claude Code model configuration. Anthropic describes Sonnet as the practical default for most coding work. |
| Opus 4.6 | Versioned Opus option available in Claude Code where your setup supports it. Opus uses more quota than Sonnet; switch when deeper reasoning is worth the cost. |
| Opus 4.5 | Dated Opus 4.5 model identifier published for Claude Code configuration. Anthropic publishes claude-opus-4-5-20251101 for environment-variable configuration. |
| Haiku 4.5 | Fast, lower-cost Claude option for quick mechanical work. Anthropic recommends Haiku for simple edits, lookups, boilerplate, and high-volume scripted runs. |
| Sonnet 4.5 | Dated Sonnet 4.5 model identifier published for Claude Code configuration. Keep this as a versioned option only when your Claude Code setup exposes it. |
Codex4 models, starting with GPT-5.5
| Model | Guidance |
|---|---|
| GPT-5.5 | Recommended starting point for most Codex tasks. OpenAI describes it as the strongest option for complex coding and research workflows. |
| GPT-5.4 | Frontier option for professional coding and tool-use workflows. Use when your account and Codex setup expose it. |
| GPT-5.4 mini | Faster, lower-cost option for lighter coding tasks and subagents. Good when latency and cost matter more than maximum capability. |
| GPT-5.3 Codex Spark | Research preview option for near-instant coding iteration. OpenAI documents gpt-5.3-codex-spark as available to ChatGPT Pro users. |
Cursor2 models, starting with Composer 2.5
| Model | Guidance |
|---|---|
| Composer 2.5 | Cursor's agentic coding model for sustained software tasks. Cursor documents composer-2.5 as the model ID for Composer 2.5. |
| Composer 2.5 Fast | Default fast variant of Composer 2.5 for interactive sessions. Cursor documents Fast as a variant of Composer 2.5, not a separate published model ID. |
Usage and billing owner
Conductor is free today. The selected provider controls model usage, limits, and billing.
| Owner | Usage path | Billing note |
|---|---|---|
| Conductor | Free today | Conductor does not bill, meter, or resell model usage. |
| Claude Code | Claude subscription or API/provider billing | The active Claude Code auth path decides which pool is used. |
| Codex | OpenAI subscription, credits, or API key | OpenAI controls account access, model availability, and rates. |
| Cursor | Cursor plan/API usage pools | Cursor controls plan usage, API usage, and on-demand billing. |
Controls and limits
Provider auth and model access decide what the harness can do. Conductor adds workspace orchestration around the session.
CURSOR_API_KEY in Settings -> Providers -> Cursor, or Conductor environment settings.Choosing a harness
Start with the provider agent whose model and workflow you already trust. Conductor adds workspace isolation, setup scripts, run scripts, terminal state, diff review, checks, pull request flow, and archive flow around that harness.
Use separate workspaces when tasks should become separate branches or pull requests. Use multiple chats in one workspace only when Claude Code, Codex, or Cursor need the same branch and current code state.
For session controls, see Agent modes. For auth and provider configuration, see Configure model providers.