OpenAI halved ChatGPT Pro $200: what Codex is worth after October 30

Updated October 9, 2026 · first published October 9, 2026

On October 30, 2026, the Codex allowance on ChatGPT Pro $200 drops from 20 times Plus to 10 times Plus. OpenAI announced the change at DevDay on September 29. The price stays at $200, so the plan now gives you half the Codex it did. On our published projection, a Pro $200 week falls from about $1,340 of API-priced usage to about $670. On the $100 tier we measured, the weekly window scaled at about half its label, so the real figure may be closer to $350.

The short version:

• What changed? Codex and ChatGPT Work usage on Pro $200 falls from 20x to 10x Plus from October 30. GPT-6 Pro messages fall from 200 to 100 a week.
• Who is protected? Existing subscribers keep the old limits until October 29. They also get a one-time $2,500 credit grant that expires at the end of the year.
• Is $200 still better value than $100? Not per unit any more. Pro $100 is 5x for $100, and Pro $200 is now 10x for $200. Both price one Plus-worth of Codex at $20.
• Does Codex still beat Claude Max 20x? On the projection, Codex still finishes about 1.2x more coding tasks, down from 2.4x. If the weekly window scales as it did on our $100 plan, Claude Max 20x comes out ahead.

What exactly did OpenAI change?

At DevDay 2026 OpenAI rebuilt its Pro line into three tiers, at $100, $200 and $500. The $200 tier lost half its Codex and Work allowance. The new $500 tier keeps a large allowance and is the only plan with Ultrafast mode, which runs at up to 300 tokens a second in Codex. Codex lead Tibo Sottiaux tied the cut to compute demand. OpenAI also said it would not bring the 5-hour limit back to Pro $200. Its pricing page currently says Pro plans have no five-hour limit, only a weekly one.

Ultrafast is less generous than it sounds. OpenAI's pricing page says Ultrafast draws on included subscription usage at 8 times the Standard rate for the same model, and Fast at 2.5 times. On Pro $500, Ultrafast buys speed by spending the allowance eight times as fast.

How much Codex does Pro $200 buy after the cut?

About $670 a week at API prices on the projection, and possibly about $350. We price a weekly window by dividing the API list-price cost of the requests by the percentage points they used, then multiplying by 100. A full ChatGPT Plus week measured a median $67 across 27 windows since August, and that is the baseline below.

The last column comes from our own logs. Between September 27 and October 8 we drained six full weekly windows on Pro $100, and they priced at a median $176, between $100 and $234. That is 2.6 times a Plus week, not 5 times. If Pro $200 scales the same way, its 10x becomes about 5.3x in practice, or about $350 a week. That is still about $1,500 of API-priced Codex a month for $200, or 7.5 times the fee. It is a good deal, just much smaller than it was. The full Pro $100 log is here.

One forum user reports the same thing from the other direction. After moving up from Pro $100, they said Pro $200 gave them roughly half the effective Codex usage they had before. Usage multipliers on these plans have not held up when people measure them.

Is Pro $200 still worth more than Pro $100?

Per unit, no. Before the cut, $200 bought 20 Plus-weeks of Codex at $10 each, and $100 bought 5 at $20 each, so the bigger plan was half price per unit. From October 30 both tiers charge about $20 per Plus-worth. hwbusters made the same point in its critique of the change. What the extra $100 now buys is headroom. You get one weekly window twice as large, instead of having to stretch a smaller one.

Here is what that means in practice:

• You use less than one Pro $100 window a week: stay on $100. Upgrading only buys space you will not use.
• You run out of a Pro $100 window mid-week: $200 doubles your room at the same unit price, so the upgrade is now a straight trade of money for capacity. Before October 30 it was a discount.
• You are an existing Pro $200 subscriber: work through your heaviest jobs before October 29, and spend the $2,500 credit grant before it expires at year end. After that, check whether one $100 window covers your week.
• You want Ultrafast: only Pro $500 has it, and it spends quota 8 times as fast. Use it for latency-bound loops, not bulk work.

Does Codex still beat Claude Max 20x at $200?

Only narrowly, and only on the projection. In our 60-day comparison, Claude Max 20x held about $1.86k of API value per week. GPT-6 Sol needed 3.3 to 4.2 times fewer dollars than Claude to finish the same coding task. Multiply the Codex dollars by that efficiency and you get the work each plan finishes in a week:

Our earlier study said Codex keeps its lead as long as the Pro $200 weekly multiplier stays above about 7x. The label is now 10x, so the projection still favours Codex. But our own $100-tier measurement ran at about half its label, and on that number Codex falls below 7x and Claude Max 20x finishes more work for the same $200. Watch your own weekly gauge before you switch. On the new Pro $200, a full Plus-sized week of work should use about 10 points of the gauge. If it uses closer to 20, you are on the lower row.

Why did OpenAI cut it?

OpenAI says it is about compute. The plans give away far more than they charge for. Even at our lower measured figure, Pro $100 delivered about $760 of API-priced Codex a month, which is 7.6 times its price. Heavy users also stack resets on top of that. OpenAI has handed out dozens of one-off weekly resets since 2025, and two global resets landed in the first week of October alone (see our Codex reset log). Cutting the $200 tier and adding a $500 tier moves the heaviest users onto a plan that charges closer to what they consume.

How to check what your plan really gives you

1. Codex writes your weekly used_percent into every session log under ~/.codex/sessions/, next to the token counts of each request.
2. Price each request at the API list price for its model. Cached input is much cheaper than uncached input.
3. For one weekly window, divide the dollars by the percentage points used and multiply by 100. That figure is what your plan's week is worth in API terms.
4. Compare it with $67 times your plan's label. If you get about half, you are seeing what we saw.

New: 12 days of real Codex Pro $100 logs

We logged every token of a heavy Codex fortnight and published the data. A full Pro $100 week was worth $175 at API prices, so the real multiplier is 2.6x Plus, not 5x. Using global and banked resets we got 8 weekly windows in 12 days, worth $1,211 of API usage. We also worked out when to redeem a banked reset and why 98% of the input was cached. Every article ships its CSV.

Sources

• OpenAI: Codex plans and usage limits (Ultrafast 8x, no five-hour limit on Pro)
• The Next Web: OpenAI cuts Pro $200 usage and launches a $500 plan
• PCWorld: ChatGPT Pro's biggest perk just got cut in half
• Hardware Busters: OpenAI cuts ChatGPT Pro $200 in half
• daily.dev: OpenAI halves Pro $200 usage at DevDay
• The Next Web: the $100 Pro plan launch
• OpenAI Community: Pro $200 giving roughly half the effective usage of Pro $100
• Developers Digest: Codex usage limits and pricing in 2026

Related

• ChatGPT Pro $100 is really 2.6x Plus, not 5x
• How resets turned a $40 plan into $1,211 of usage
• When to redeem a Codex banked reset
• 98% of our Codex input was cached: the bill
• 8 Codex weekly windows in 12 days on Pro $100
• Claude Max 20x vs ChatGPT Pro: 60 days of logs
• How much Codex ChatGPT Pro includes
• Codex reset log
• Claude Max 20x vs 5x: the weekly limit, measured
• How many tokens Claude Pro and Max give you

Want this applied to your stack? Bring the provider bills, gateway logs, and top workflows; we will map the cost drivers and savings path. Book a free audit →

Back to finopsllm.com