Claude Pro vs ChatGPT Plus: Claude Code Gets About 3x the Tokens of Codex

Claude Pro and ChatGPT Plus cost the same $20, and both put a coding agent inside the chat subscription: Claude Code on one side, Codex on the other. That makes them easy to compare on how much each one actually lets you use.

At $20, it isn’t close. Claude Pro gives you about three times the tokens ChatGPT Plus does.

Quick Answer

• Which gives more usage, Claude Pro or ChatGPT Plus? Claude Pro. About 2.9 billion tokens a month in Claude Code against about 1 billion in Codex, per SemiAnalysis’s October 2026 limit tests.
• Which is better value per dollar? Claude, at every tier. About $57–59 of API value per $1 on Claude plans, about $10.50 on ChatGPT plans.
• Is Claude Max worth it? Only if you hit Pro’s limit every week. Max buys more room at the same rate.
• Where ChatGPT Plus still wins: published limits you can check, effectively unlimited everyday chat, and an $8 entry plan.

What this isn’t: a quality benchmark. It measures how much each plan lets you use, not how good the code is. For the record, Opus 5.5 sits at #1 on both Arena’s agent leaderboard and OpenRouter’s benchmark ranking as of October 24.

Claude Pro vs ChatGPT Plus Usage Limits, in Tokens

Claude on Opus 5.5, ChatGPT on GPT-6.1 Sol, both on an agent workload. Tokens means everything: input, cached input, cache writes and output.

Claude Code vs Codex: API Value per Month

“API-list value” is what the same tokens would cost on the API. SemiAnalysis measured how far each plan’s usage meter moves per million tokens, worked out how many tokens a full month’s limit holds on an agent workload, and priced those tokens at each company’s own API rates.

At $20, Claude Pro is worth about $1,178 of API time. ChatGPT Plus is worth $211.

Read the fine print before quoting these:

• The dollar gap overstates the token gap. GPT-6.1 Sol is much cheaper per token on the API than Opus 5.5, so the value lead (~5.5x) is bigger than the token lead (~2.8x). The token number is the fairer one to quote.
• The two sides don’t measure the same thing. SemiAnalysis measured Claude through Claude Code and ChatGPT through Codex. Claude’s chat draws on the same limit; ChatGPT Plus chat is effectively unlimited on top of Codex.
• The workload is cache-heavy. SemiAnalysis’s agent mix is 96.6% cached input. A workload with more fresh input or output would land on different numbers.
• ChatGPT’s $200 plan just got halved. OpenAI cut its token limits in half a week before SemiAnalysis published, and added a $500 tier. The $200 figures above are after the cut. If you bought Pro $200 before the cut, you keep the old limits (twice the tokens per model) until October 29. After that, every $200 plan is on the halved limits shown here.
• This is allowance, not work done. SemiAnalysis puts realistic average use at about 20% of the limit. I use Claude a lot, and I’ve never hit the weekly limit in two days (fair warning: I’m a moderate, everyday user, not someone running coding agents all day).

Which AI Subscription Gives the Most Usage per Dollar?

API-list value ÷ monthly price:

• Every Claude tier beats every ChatGPT tier, by about 5.5x. The cheapest Claude plan gives more value per dollar than the most expensive ChatGPT plan.
• Upgrading doesn’t get you a discount on either side. Max 5x is slightly worse per dollar than Pro. Max 20x is level with it. You pay for more room at the same rate.

Is Claude Max Worth It Over Claude Pro?

Allowance only matters if the job actually ends, the same logic behind the cheapest model that finishes the job. Most people believe they live at the limit right up until they check their usage page for a week.

Is Codex Cheaper Than Claude Code?

On sticker price, yes. ChatGPT Go ($8) runs Codex, but on GPT-6 Luna only. If you want Codex on Astra or Sol, you need Plus at $20, and at $20 Claude Pro gives you more. For plans under $20, see the cheapest AI coding plans, priced per dollar.

Do Claude Pro and ChatGPT Plus Have 5-Hour Limits?

Both do, plus a weekly limit. The difference is what each company tells you.

• ChatGPT Plus publishes its Codex limits as ranges per model. 5–45 local messages per five hours on GPT-6 Astra, 15–160 on GPT-6.1 Sol, and 350–3,000 on GPT-6 Luna. OpenAI says these are estimates, not fixed limits. ChatGPT’s Pro plans drop the 5-hour window and keep only the weekly one.
• Claude Pro doesn’t publish numbers at all. You get a five-hour session limit, a weekly limit, the 5x and 20x multipliers between tiers, and a usage page.

Claude gives you more and tells you less.

You can check Codex’s limits before you pay; Claude’s you find by hitting them.

How This Was Checked

• Sources: SemiAnalysis’s subscription-value and token figures (October 6, 2026). Plan prices and limits from chatgpt.com/codex/pricing and claude.com/pricing; plan catalog behind our AI coding plan guide. Leaderboards read October 24, 2026.
• Token counts are SemiAnalysis’s, for Claude on Opus 5.5 and ChatGPT on GPT-6.1 Sol.
• Value per $1 = API-list value ÷ plan price, our layman math.
• Codex message ranges are OpenAI’s published estimates. Claude Max API credits per Anthropic’s pricing page and its October 2026 announcement.
• Checked against the report: every dollar and token figure above matches SemiAnalysis’s own charts. We didn’t rerun their meter tests. No head-to-head quality testing. Limits change often; confirm at checkout.

For Claude Code-heavy work at $20, buy Claude Pro. Buy ChatGPT Plus if you want unlimited chat and limits you can read before you pay.