Codex vs Claude Code: The Cost Structure Comparison

Codex and Claude Code get compared on model quality everywhere, so this comparison covers the part that decides budgets: cost structure. Both sell flat subscriptions for interactive coding. They split hard on programmatic use. Codex automation bills against the same flat ChatGPT plan windows, by OpenAI’s design. Claude Code automation meters against a monthly Agent SDK credit and then moves to per-token API billing, and Anthropic prohibits third-party services from running on subscription credentials at all.

What each subscription includes

Codex is included in ChatGPT Free, Go, Plus, Pro, Business, Edu, and Enterprise plans, and OpenAI’s README recommends ChatGPT sign-in (github.com/openai/codex). Capacity arrives as usage windows: as planning estimates, Plus at $20 absorbs roughly $700 of API-equivalent work a month, Pro 5x at $100 roughly $3,500, Pro 20x at $200 roughly $14,000. Estimates, never guarantees; OpenAI publishes and adjusts the actual limits on its pricing page.

Claude Code is included with Claude Pro ($20) and Max plans ($100 for 5x, $200 for 20x), also delivered as usage windows for interactive sessions. Since June 2026, those plans additionally carry a metered Agent SDK credit, sized to the tier: $20 a month on Pro, $100 on Max 5x, $200 on Max 20x, covering personal programmatic use such as claude -p and the Agent SDK in your own projects.

That last sentence contains the structural split. On the OpenAI side, programmatic use draws from the same windows as interactive use. On the Anthropic side, programmatic use is a separate, metered allowance.

The programmatic-use policy table

Question	Codex (OpenAI)	Claude Code (Anthropic)
Documented non-interactive mode	codex exec, for scripts and CI (developers.openai.com/codex)	claude -p, for personal use against the Agent SDK credit
Plan-backed automation	Draws on plan windows	Draws on the metered credit, then API billing
Headless sign-in	Device-code flow, documented	Subscription OAuth restricted to Anthropic’s own apps
Third-party harnesses on plan login	Sign in with ChatGPT exists; account rules apply	Not permitted on subscription credentials
Hosted services on plan credentials	Not specifically addressed; OpenAI has the final call	Expressly prohibited; enforced January 2026 against OpenCode and others
Overflow when limits hit	API key, per token	API key, per token

Anthropic’s docs state it does not permit third-party developers to offer Claude.ai login or route requests through Free, Pro, or Max plan credentials on behalf of their users (code.claude.com/docs/en/legal-and-compliance). The June 2026 credits restored personal programmatic use; the hosted-service shape stayed banned. The full history, including the January 2026 enforcement, is in why we don’t support Claude Code.

The math at three usage shapes

Interactive only. A developer in the terminal all day fits either vendor’s flat plan. Costs are comparable at each tier and the deciding factors are model preference and how generous each vendor’s windows feel for your usage, which both vendors tune over time.

Personal automation, $1,000 a month of API-equivalent volume. On the Codex side, that volume fits inside a Pro 5x window estimate at $100 flat, or $229 all-in through a managed endpoint. On the Claude side, a Max 5x plan contributes its $100 credit and the remaining roughly $900 bills at Anthropic’s published API rates (anthropic.com/pricing), so total cost tracks volume. Windows absorb volume; credits meter it. That one sentence is the comparison.

Team or hosted automation. On the Codex side, plan-backed automation through third-party harnesses uses documented sign-in flows under OpenAI’s account rules: intended functionality, with OpenAI keeping the final call. On the Claude side, this shape is prohibited on subscription credentials regardless of volume, so the price is the API price. For a team burning $3,500 of monthly volume, that is roughly $229 all-in versus roughly $3,500, with the caveat that the Codex figure is a capacity estimate, not a guarantee.

The migration sentiment

The cost-structure gap shows up in community sentiment. As one heavy user put it in a developer forum: “I get way more out of the codex $100 plan than I was getting out of the Anthropic $200.” That is one data point about perceived capacity per dollar, not a benchmark, and window generosity shifts as vendors retune limits.

The fair framing: Anthropic chose to sell programmatic Claude through the API and meter the subscription path. OpenAI chose to include programmatic Codex inside plan windows. Neither is an accident; they are different answers to the same business question, and the buyer’s job is to price their own workload under each answer. The Plus-tier version of that arithmetic is in ChatGPT Plus vs the API for coding.

Both on the API

Strip the subscriptions away and the structures converge: both vendors meter per token. GPT-5 lists at $1.25 per million input tokens and $10 per million output as of June 2026 (openai.com/api/pricing); Anthropic publishes its per-model rates at anthropic.com/pricing. On a meter, agent-style coding workloads bill every plan step, tool call, and retry, so costs scale with loop count on either vendor. The API is the right lane for spiky workloads and for anything a subscription’s terms exclude.

What this means if you automate

If your coding automation runs on Claude, your costs are the credit plus the meter, by Anthropic’s design, and the details of that system are in Anthropic’s Agent SDK credits, explained. If it runs on OpenAI models, plan windows can carry the bulk of it: Codex Hosted runs OpenAI’s official Codex CLI signed into your own ChatGPT account, $129 a month with no inference markup, complete responses on the Codex lane, and your API key as overflow.

The calculator prices your current metered spend against the subscription-backed structure in about thirty seconds.