Codex vs Claude Code Pricing, Claude Code vs Codex Cost for Teams

The short answer: Codex and Claude Code cost the same at almost every level a US team buys. Both are $20 a month for one developer (Claude drops to $17 on annual billing), both charge $20 per team seat billed annually or $25 monthly, and both sell a $100 heavy-use seat. Where they split is the top tier, which OpenAI stopped selling to new subscribers on September 10, 2026, and the token rates behind the most capable model on each side.

Most comparisons ranking for this search spend their time on benchmarks and describe pricing from last spring. Since then OpenAI moved Codex to token-based credits, added GPT-6 Sol and GPT-6 Luna on September 22 and paused its $200 Pro tier. Anthropic published its Max 20x price and kept Claude Code in every paid Claude plan. Everything below is read from OpenAI's Codex pricing doc and rate cards and from Anthropic's pricing page and help center in late September 2026. For the full plan detail on each side, see our Codex pricing and Claude Code pricing breakdowns.

Codex vs Claude Code pricing, plan by plan

Two rows deserve a second look. The heaviest-individual row is the only place the vendors genuinely diverge on list price: a developer who wants 20 times the base allowance today can buy it from Anthropic and cannot from OpenAI unless they already hold it. And the Enterprise row is the only place one vendor publishes and the other does not. Anthropic prints a $20 Enterprise seat with usage billed at API rates; OpenAI quotes Enterprise through sales.

What the team seat actually buys on each side

The team seats cost the same, but they are measured in different units, so do not read the matching $20 as matching capacity. OpenAI says a Business Standard seat carries the same estimates as Plus, which on GPT-6 Sol is 15 to 150 messages per five-hour window. Anthropic says a Team Standard seat is 1.25 times Pro per session, with a weekly limit on top that it does not publish as a multiple. Neither vendor gives a fixed message count, and both say ordinary chat and coding draw from one shared pool.

The heavy seats line up more closely. OpenAI's $100 Business seat "uses the Pro 5x estimates," so it is five times a Standard seat. Anthropic's Team Premium seat is five times a Team Standard seat. Same price, same ratio to the base seat. The practical difference is monthly billing: Anthropic publishes $125 for a monthly Premium seat, and OpenAI's Codex pricing doc lists the $100 figure only.

For a 10-developer US team, list prices come out like this:

The seat bill is a tie to the dollar. The last row is the one that decides the budget, because both vendors are telling you that a working developer consumes several times what a $20 seat costs. That extra consumption is paid either with heavier seats or with usage billed past the allowance, which is where the two products stop looking alike.

Where the bills diverge on token rates past the seat

When a developer runs out of plan allowance, both vendors let work continue on metered usage. OpenAI bills Codex in credits worth $0.04 each; Anthropic bills usage credits at API rates. Comparing the models each vendor recommends for everyday coding and for the hardest work:

The everyday tier is identical: GPT-6 Sol and Claude Sonnet 5.5 (like Sonnet 5 before it) both cost $2 in and $10 out per million tokens. The top tier is not. GPT-6 Astra costs two and a half times Claude Opus 5.5 per token, so a team that routes its hardest refactors to the frontier model pays noticeably more per task on Codex, before counting how many tokens each model spends. At the cheap end the two now match: GPT-6 Luna and Claude Haiku 5.5 both cost $0.10 in and $0.50 out (Haiku 5.5 rises to $0.50 and $2.50 on prompts over 100,000 tokens), a tenth of the older Haiku 4.5 at $1 and $5. Luna is also the reason Codex can offer 350 to 3,000 Luna messages per window on a $20 seat.

Two OpenAI multipliers are worth budgeting for. Fast mode uses 2.5 times the Standard credit rate on GPT-6 models, and cloud chats on ChatGPT plans run on GPT-5.6 Sol, which costs twice GPT-6 Sol per token. Neither is a reason to avoid Codex, but both are settings someone should own. On the Claude side, the equivalent discipline is setting spend limits on usage credits before turning them on, which Team and Enterprise admins can do at the organization, group or member level.

Which one should you buy?

Buy on the model you want as your default, because list prices will not separate them. If your developers mostly use the everyday model and want the cheapest possible tier for simple edits, Codex with GPT-6 Sol as the default and Luna for small tasks is hard to beat per token. If they lean on the frontier model for hard problems, Claude Code with Opus 5.5 costs less per token than GPT-6 Astra, and Max 20x remains available for the one or two people who live in the tool.

• Solo developer, part time: either at $20. Claude Pro annual saves $36 a year.
• Solo developer, all day: Claude Max 20x if you need the top tier today; otherwise either $100 plan.
• Team of 5 to 50: either team seat at $20 annual, heavy seats for the few who hit limits, and a written cap on metered usage.
• Enterprise with a procurement process: Anthropic publishes its Enterprise seat price, which shortens the budgeting step; OpenAI's is negotiated. Our Claude Enterprise pricing and ChatGPT Enterprise pricing pages cover both contracts.

Whichever you pick, the seat is rarely what hurts. The overspend comes from people and process: two tools licensed for the same developer, heavy seats handed out by default, or a backlog nobody has the hands to ship. If the real bottleneck is engineering capacity rather than tokens, it can be cheaper to hire AI developers for a fixed-scope build than to put every engineer on a $100 seat.

The other lever is taking repeated work off the metered tools entirely. Reviewing every pull request against your conventions, triaging bug reports and keeping docs current all burn allowance on either product. A ready-made agent that owns one of those jobs runs at a flat monthly price with no messages or tokens to count: see Code Reviewer, Bug Triager and Docs Writer. And if Cursor is on your shortlist too, our Claude Code vs Cursor pricing comparison runs the same math against Cursor's seats.