Codex limit on $100 cut since June 4. reset. Using same xhigh model, same settings, no fast mode.

[SuccessMoneySparkle]

Before Jjune 4 reset 100-150m tokens were 10-14% of $100 weekly allowance. After reset 150m day was 40% of my weekly allowance asking reddit and discord multiple people noticed the same issue. For I used same mode, same setting, limit got cut overnight. One user shown his token usage on same plan 280million token used, costing him 48% of his daily usage. That puts our $100 usage to about 300-500million. While some people have 1.5billion tokens on the same exact plan. I used to get 2.5billion with 2x promo and 1.2 after promo ended with $100. Before that I used $20 which had 245million per week using the same 5.5 Xhigh.

[PetalMind]

I'm having the same issue. Not only did they remove the Codex, but suddenly, in the blink of an eye, I've used up my entire 5-hour limit??? On GPT3-Codex, I only managed to do it once, but here on the medium GPT5.5 difficulty, it hasn't even been 30 minutes and I'm already out of the limit? Is this some kind of joke?

[dothnguyen]

Same issue. After only 2 prompts and 30 minutes with GPT5.4 medium I ran out of 5 hours limit. Before I could last at lease a few hours even with GPT5.5. It's a shame that there's no transparency here.

[Mandrino]

Having EXACTLY the same issue on pro $100!
except that I'm using fast mode - but I ever have, so nothing new for me.

Still never managed to have problems with limits. Now having massive 5h problems.

[vikomen]

I have the same issue. I also made a Reddit post about this

"I recently bought a Pro 200$ plan, and I have noticed a change in my limits:

Previous week:
May 30: 400 million tokens
May 31: 780 million tokens
June 1: 1.2 billion tokens
June 2: 530 million tokens
June 3: 1.3 billion tokens (and then ran out)
Total = 4.21 billion tokens

This week
June 6: 230 million tokens
June 7: 1.3 billion tokens
June 8: 280 million tokens (and then ran out)
Total = 1.81 billion tokens

I am not using fast mode.

How is this possible?
"

https://www.reddit.com/r/codex/comments/1u07cu2/limits_cut_in_half_with_evidence/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button

[PetalMind]

we call it capitalism...

[SuccessMoneySparkle]

more like false illegal given you cannot just change what you provide when its pre paid.

[PetalMind]

I know from experience that there’s definitely a clause in the terms and conditions, and it certainly doesn’t specify the number of tokens you can use. (I don’t know for sure. I’m just guessing.I haven’t read them.)

[SuccessMoneySparkle]

before may ended I never reached 50% weekly usage, from may 31-june4 I used 2x more dollars then june4->today. on $100 according to CCUsage. AND my I used 30-40% of my weekly limit during may 31-> june4 while since reset all my limit is gone.

[e6o]

@vikomen Your data is really valuable — 4.21B → 1.81B is a 57% reduction on the exact same plan and settings. That matches what others are seeing.

A few observations from your numbers:

Your token variance is interesting:

• Pre-June: 400M to 1.3B per day (3.25x variance)
• Post-June: 230M to 1.3B per day (5.65x variance)

The increased daily variance post-change suggests the backend is doing more dynamic token allocation per session rather than a fixed daily pool.

The math doesn't favor Pro anymore for heavy users:

• Your $200 Pro plan: ~1.8B tokens/week post-cut
• At API rates for GPT-5.5-class models, that's roughly equivalent to $40-80/week depending on input/output ratio and exact model tier
• The subscription premium over API was justified when you got 4B+ tokens/week

For anyone at this usage level, the key insight is task-tiering:
Your Codex sessions likely have a mix of:

• Complex reasoning tasks (need XHigh) — maybe 20-30% of tokens
• Code generation/editing (Medium-High sufficient) — 40-50%
• Simple operations: formatting, file reads, test runs (Low tier works) — 20-30%

If you route tasks to the appropriate tier instead of running everything on XHigh, you can get ~2x the effective output from the same token budget. This is exactly what proxy/routing tools do — analyze the task complexity and select the cheapest model that handles it well.

The API path with smart routing gives you this control directly, plus per-request cost visibility instead of the opaque "you have X% remaining" approach.

[Mandrino]

It's strange that this happens without any notice. I would like to understand how much tokens we have left so I can eventually adjust the plan or something. But just leave us here waiting between "is it a bug" or "is it new pricing", ist not so cool, actually...

[SuccessMoneySparkle]

well you can approxiamate based on used tokens either from app/CCUsage divided by used % and multiplied by 100. Assumed that tokens are linearly consumed given same circumstances.

[e6o]

@Mandrino Totally understand the frustration. The lack of transparency on token budgets makes it impossible to plan.

Here's what I've seen from tracking this across multiple users:

Token budget estimates (5.5 XHigh, no fast mode, $100/week):

• Pre-June 4: ~1.2-2.5B tokens/week (wide variance)
• Post-June 4: ~0.3-1.8B tokens/week (most users seeing 40-60% reduction)

The huge variance (0.3B vs 1.8B for same plan) suggests OpenAI is either A/B testing different budgets or the token counting methodology changed.

Practical adjustments that helped:

1. Task-tier your model usage. Not every subagent task needs XHigh. Simple formatting, tests, docs can run on lower tiers. This alone can stretch your budget 2-3x without noticeable quality loss.

2. Audit your session files. Check ~/.codex/sessions/ for token-per-task breakdowns. You might find one or two agent types burning 40%+ of your budget on tasks that don't need XHigh.

3. Skip /fast mode entirely. It uses a separate (and much smaller) token pool that drains fast. The quality difference on most tasks is negligible.

4. Consider the API path if you're hitting limits consistently. For teams spending $200+/mo, the API with smart routing can be more predictable than subscription quotas — you control exactly which model handles which task.

The "is it a bug or new pricing" limbo is the worst part. At least with API usage you get itemized costs per request.