OpenAI has quietly adjusted Codex's pricing logic. The changes are not dramatic, but the direction is telling.
Previously, using Codex within ChatGPT Business and Enterprise was essentially a per-seat monthly subscription. Now there is an additional option: usage-based billing for dedicated Codex seats, allowing teams to pay based on actual consumption without committing to a fixed monthly fee. At the same time, the standard ChatGPT Business subscription fee has dropped from $25 per seat to $20 per seat (annual billing).
Individually, none of these announcements are explosive. Taken together, they send a clear signal: AI coding tools are moving from a "trial phase" into a serious commercialization stage.
What's New in Codex Itself
Beyond pricing, there are also substantial product-level improvements.
Plugin system upgraded to a first-class citizen
Previously, Codex's plugin support was somewhat scattered. Now it is unified—workspace plugins are automatically synced on startup, the /plugins command allows direct browsing, and the install/uninstall process has been standardized.
For developers, this means Codex can be truly embedded into internal company toolchains, rather than being an isolated coding assistant.
Address system for multi-agent workflows
This change is a bit technical: previously, when multiple agents called each other, they used opaque IDs. Now, they use readable path formats—such as /root/agent_a or /root/agent_b.
It sounds like a minor optimization, but the impact is significant. Path-format addresses make relationships between agents easier to trace and debug. When something goes wrong, you know where to look. The engineering complexity of multi-agent coordination has been reduced.
Enterprise Plugin Directory
Business and Enterprise versions now have a unified plugin directory, and administrators can control which plugins are available to employees. This feature is important for large companies—security audits and compliance controls now have a formal entry point.
Who Benefits from Usage-Based Billing
A rough calculation:
- Small teams with light usage: If each person runs only a few coding tasks per month, a monthly subscription might still be cheaper. The new plan offers average value here.
- Heavy-usage engineers: For those who use Codex deeply every day, a monthly subscription is more predictable. However, the new plan includes a $100 free credit per person (up to $500 per team), which can save a lot in the first few months.
- Project-based teams with fluctuating usage: Usage-based billing is clearly more suitable here—pay more during peak project periods and less during slow periods.
OpenAI is also offering a promotion: eligible workspaces can receive up to $100 in Codex credits for each new user added, with a total cap of $500.
The Next Phase of AI Coding Tool Competition
Over the past year, the AI coding tool space has become increasingly competitive:
- GitHub Copilot continues to iterate, adding an Agent mode this year.
- Cursor/Windsurf has a large base of loyal users in the standalone IDE direction.
- Claude Code (Anthropic) recently faced developer backlash due to third-party tool restrictions.
- On the open-source front, local alternatives are maturing quickly.
In this context, Codex's latest adjustments are not about making a big splash in performance. Instead, the focus is on compliance and manageability for enterprise deployment—administrator control over the Plugin Directory and traceable paths for multi-agent workflows are both aimed at large customer needs.
In cost-benefit decisions, pure technical capability is only part of the equation. Whether an enterprise trusts the tool and can manage it effectively is increasingly important.
From this perspective, OpenAI's update is a pragmatic business decision, not a technological breakthrough—but pragmatic business decisions can sometimes determine market share more than technological breakthroughs.
Sources: CocoLoop, OpenAI Release Notes - April 2026 Latest Updates (Releasebot), Introducing Codex (OpenAI)