Availability And Rollout Status

Issue 75 Edition 2026-03-16 5 min read

Not accepted General

Sources: 1 • Confidence: Medium • Updated: 2026-04-12 10:16

Key takeaways

OpenAI Codex subagents are generally available after several weeks of preview behind a feature flag.
Custom Codex agents can include custom instructions and can be pinned to specific models, including gpt-5.3-codex-spark for speed.
Custom Codex agents can be referenced by name in prompts to orchestrate multi-step workflows where different agents reproduce bugs, trace code paths, and implement fixes.
Available information does not clearly explain the distinction between the worker and default Codex subagents.
The subagents pattern is supported across multiple coding-agent platforms, including Codex, Claude Code, Gemini CLI, Mistral Vibe, OpenCode, Visual Studio Code, and Cursor.

OpenAI Codex subagents are generally available after several weeks of preview behind a feature flag.

Custom Codex agents can include custom instructions and can be pinned to specific models, including gpt-5.3-codex-spark for speed.

Custom Codex agents can be referenced by name in prompts to orchestrate multi-step workflows where different agents reproduce bugs, trace code paths, and implement fixes.

Available information does not clearly explain the distinction between the worker and default Codex subagents.

The subagents pattern is supported across multiple coding-agent platforms, including Codex, Claude Code, Gemini CLI, Mistral Vibe, OpenCode, Visual Studio Code, and Cursor.

What are the exact behavioral differences and intended use cases for worker subagents versus default subagents in Codex?
What eligibility, rollout conditions, or account limitations apply to the general availability of Codex subagents (e.g., is the feature flag fully removed for all users)?
What are the pricing, quota, or rate-limit implications of running multiple subagents in parallel within Codex?
What are the concrete performance characteristics (latency, reliability) and tradeoffs of gpt-5.3-codex-spark relative to other pin-able models for subagents?
How much feature parity exists across the listed subagent-supporting platforms (capabilities, orchestration primitives, agent isolation, tool access)?

General availability of Codex subagents could reduce gating friction and expand adoption surface for multi agent coding workflows.
Per agent model pinning suggests a packaging approach where users trade off speed versus capability by role, potentially increasing usage through more predictable workflow tuning.
Named subagent orchestration for debugging indicates movement toward modular agent workflows rather than single chat sessions, which may increase tool stickiness if it improves developer throughput.

Clear documentation or release notes stating subagents are default on for all eligible users, with removed or minimized feature flag gating.
Published details on pricing, quotas, or rate limits for parallel subagent runs, and performance characteristics of gpt-5.3-codex-spark versus other pinnable models.
Clarified semantics and examples distinguishing worker versus default subagents, plus evidence of feature parity and reliability across other subagent supporting platforms.

GA still requires feature flags, invites, or significant account limitations that keep adoption constrained or inconsistent across users.
Parallel subagents incur restrictive quotas or materially higher costs that limit practical multi agent orchestration.
Worker versus default subagent ambiguity remains unresolved, leading to misuse or poor team enablement and reducing perceived value of the subagents pattern.