Daily-It

Claude Code and OpenAI Codex: Why Might AI Coding Agent Pricing Be Splitting?

AI coding agent pricing is no longer easy to explain as a single flat subscription. Claude Code and OpenAI Codex both position themselves as developer-facing AI agents, but they reveal different instincts about where subscriptions end and metered automation begins.

Summary

A Korean Brunch post titled “Claude pricing changes from June 15” captured an important signal: Anthropic had planned to separate Claude Agent SDK usage from ordinary Claude subscription usage. However, Anthropic’s official Help Center now says that the June 15 change has been paused. So the accurate reading is not “the split already happened,” but rather “Anthropic tried to draw the line, then paused before it took effect.”

OpenAI Codex, by contrast, is currently presented in official docs as included in ChatGPT Free, Go, Plus, Pro, Business, Edu, and Enterprise plans. It is not unlimited: plan limits, ChatGPT credits, and API-key based token billing still exist. But the product message is different. OpenAI is making Codex feel like a core part of the ChatGPT subscription.

Table of Contents

What this article covers

  1. Why this pricing question matters
  2. What Anthropic planned for Claude Agent SDK
  3. Why the June 15 change needs careful wording
  4. How Claude Code pricing works now
  5. How OpenAI Codex looks different
  6. The strategic difference
  7. What users should check
  8. FAQ
  9. Conclusion
  10. References

Why this pricing question matters

AI subscriptions used to be fairly simple. A person sat in front of a screen, typed a prompt, waited for an answer, and continued the conversation. In that world, the user’s cost expectation was close to “how much can I use for a fixed monthly fee?”

AI coding tools have changed the shape of usage. They can modify code in a terminal, review changes from an IDE, run inside GitHub Actions, or be called by third-party apps while the user is away. The human may be asleep, but the agent may still be spending tokens.

AI that a person uses directly and AI that keeps working in the background are difficult to price in exactly the same way.

That is the heart of this issue. AI companies must decide whether automated agent usage belongs inside the same subscription bucket as interactive chat, or whether it should move into credits and API-style metering.

What Anthropic planned for Claude Agent SDK

The Brunch post discussed Anthropic’s planned Claude Agent SDK pricing change. According to the preserved explanation in Anthropic’s official document, Anthropic intended to stop counting Claude Agent SDK and claude -p usage against ordinary Claude plan limits starting June 15, 2026.

Item Planned structure
Covered usage Claude Agent SDK, claude -p, Claude Code GitHub Actions, and third-party apps built on the Agent SDK
Subscription usage Claude web/app, interactive Claude Code, Claude Cowork
New bucket A separate monthly Agent SDK credit
After credit runs out Standard API rates if usage credits are enabled; otherwise requests stop

The planned monthly credits were also explicit: $20 for Pro, $100 for Max 5x, $200 for Max 20x, with Team and Enterprise amounts depending on seat type.

In other words, Anthropic was trying to separate “interactive subscription usage” from “automated background agent usage.”

Why the June 15 change needs careful wording

The important update is that Anthropic’s Help Center now includes a June 15 notice saying the change is paused. For now, Claude Agent SDK, claude -p, and third-party app usage still draw from subscription usage limits.

So it would be inaccurate to write that Claude Agent SDK pricing has already changed on June 15. The more accurate wording is: Anthropic planned the split, then paused it before it took effect.

That does not make the episode irrelevant. If anything, it shows how sensitive agent pricing is. A user may feel that automation was part of the subscription, while the provider may see long-running agent workloads as a separate cost center.

How Claude Code pricing works now

Claude Code is currently described as included with Pro and Max subscriptions. Terminal and supported IDE usage share the same limits as Claude on the web, desktop, and mobile apps.

  • If an ANTHROPIC_API_KEY environment variable is set, Claude Code can use API billing instead of subscription usage.
  • When a plan limit is reached, users can enable usage credits to continue at standard API rates.
  • Usage credits are billed separately from the Claude subscription.
  • Non-interactive automation such as GitHub Actions or claude -p may remain more sensitive to future policy changes.

So Claude Code is included in the subscription today, but there are already bridges into API-style metering depending on how it is used.

How OpenAI Codex looks different

OpenAI’s Codex pricing page presents Codex as included in ChatGPT Free, Go, Plus, Pro, Business, Edu, and Enterprise plans. Plus includes Codex on the web, CLI, IDE extension, iOS, cloud-based code review, and Slack integration. Pro offers 5x or 20x higher Codex usage than Plus.

Plan / path How OpenAI positions Codex
Free / Go Light exploration or lightweight coding tasks
Plus Web, CLI, IDE extension, iOS, and cloud integrations
Pro 5x or 20x more Codex usage than Plus
API Key CLI, SDK, or IDE extension with token-based API pricing; no cloud-only features
Business / Enterprise Seat and usage-based expansion for organizations

The message is simpler: Codex is part of ChatGPT. Limits and credits still matter, but the product is being packaged as a subscription feature first.

The strategic difference

The difference is about where each company places agent usage.

Question Anthropic Claude Code / Agent SDK OpenAI Codex
Core message Claude Code is included; Agent SDK automation was considered for a separate bucket Codex is included in ChatGPT plans
Automation Separate Agent SDK credits were planned, then paused Framed through plan inclusion, credits, and API expansion
Overage path Usage credits or API rates ChatGPT credits or API-key token billing
User feeling More explicit boundary, but also more complex More bundled and approachable, but still limited

Anthropic appears more willing to expose the cost boundary between interactive use and autonomous automation. That makes sense operationally: background agents can consume tokens unpredictably and continuously.

OpenAI appears to be using Codex as a stronger subscription value driver. The likely path is to build habits inside ChatGPT plans first, then move heavy users toward Pro, Business, credits, or API billing.

Neither company is promising truly unlimited coding agents. The difference is in packaging: Anthropic tried to reveal the automation meter more directly, while OpenAI currently makes Codex feel more bundled into the subscription.

What users should check

For Claude Code users

  • Check whether ANTHROPIC_API_KEY is set in your terminal environment.
  • Confirm that Claude Code is logged in with your subscription account if that is what you intend.
  • Use /status or /usage to monitor usage.
  • Understand that enabling usage credits may create charges outside the subscription.
  • Treat GitHub Actions, claude -p, and SDK automation as areas to watch for future policy changes.

For OpenAI Codex users

  • Check whether you are on Free, Go, Plus, Pro, Business, Edu, or Enterprise.
  • Confirm which Codex surfaces are included: web, CLI, IDE extension, iOS, code review, Slack, or API.
  • If you use an API key, treat the usage as token-based API billing rather than ordinary ChatGPT subscription usage.
  • Watch ChatGPT credits and plan limits to avoid surprise overage behavior.

FAQ

Did Claude Agent SDK pricing actually change on June 15?

Based on Anthropic’s official Help Center, no. The planned change was paused on June 15. The accurate statement is that Anthropic announced and documented the split, then paused it before it took effect.

Is Claude Code included in a subscription?

For Pro and Max users, Claude Code is included and shares plan limits with other Claude surfaces. API keys and usage credits can still move usage into separately billed territory.

Is OpenAI Codex free?

Not exactly. Codex is available across ChatGPT plans, including Free, but heavier usage and more capabilities are tied to paid plans, credits, or API billing.

Will AI coding agents become subscription-based or usage-based?

Most likely both. Light interactive usage will remain bundled into subscriptions, while long-running automation and large workloads will increasingly connect to credits or API-style metering.

Conclusion

Claude Code and OpenAI Codex have not completely split into two pricing worlds yet. Anthropic’s Agent SDK separation was paused, and OpenAI’s Codex still has limits and metered expansion paths. But the direction is clear: AI coding agents are pushing subscription pricing toward a hybrid model.

Subscriptions will remain the entry point. The more an agent works autonomously, continuously, or at organizational scale, the more likely it is to meet credits, limits, and API-style billing. That is why the better question is not whether the pricing has already split, but why it may be starting to split.

References