Codex Resets(codex-resets.com) |
Codex Resets(codex-resets.com) |
But that hasn't happened, and it may or may not ever happen; we don't know the future. All we know is the past and the present.
And that today, we have tokens to burn.
nah, I'll use open models. Only way I'd use codex now is if it was free, and I don't mean someone else paying for it. I would rather have them pay for DSv4P or Kimi
Retroactive refunds are certainly better, but I think OpenAI is much more customer-focused in this regard than Anthropic.
> 4 days ago: 8M users
> 3 days ago: 9M users
That's some incredible growth.
One person using the app on two computers and a phone? That's obviously 3 users.
Growth Hackers and SEO types have done so much to make the world an insufferable place. I congradulate them on the OpenClaw psyop, that was some very inspired bullshit that did fuck all and is now irrelevant.
Best use the AI subsidies while you can!
It is a clever growth hack.
I've been enjoying the resets, plus I had 3 resets I haven't used.
Just using 5.6 Sol in Fast mode the whole time, 1B tokens per day.
However, when they removed the 5h limit they also quietly lowered the 5.6 Sol context from 354k to 258k or something like that. I noticed it in Codex.
The only way to do it now is through shenanigans with the Codex App Server which is not ideal.
Just ask Codex to use its local auth token as a bearer token and send a GET request to it. The response includes "available_count" and "credits[].expires_at". Or script it yourself obviously.
https://chatgpt.com/backend-api/wham/rate-limit-reset-credit...
For the automated checking of other stuff: It was a one-shot prompt to get Codex clank up some Python that returns remaining usage, next reset time/date, and so on.
The result does use Codex App Server, but it's a short-lived process that is dealt with over stdio so that's... fine-ish, I guess?
If you keep staying with them, they'd feel less and less threatened.
You need to use them about half as much as you do a Chinese model so that the usage stats that get collected show them that they are still losing users (and attention) to the Chinese models.
I'm trying to squeeze what I can in the meantime but I feel like they said it was getting cut off multiple times and they keep extending and possibly resetting it (havent been looking close enough to know for sure)
No, they've been clear about the fact that Fable is staying in indefinitely now.
They also extended the 50% extra usage thing through to August 19.
I’m guessing they’re getting ready.
1. Your window ends (usage reset) in 3 days
2. They reset usage for everyone
3. Your usage goes back to 100% and your current window ends in 7 days now.
They've bought a lot of dev goodwill tho, which matters I guess.
Codex usage is clearly common enough to have entered the vernacular of folks here on HN.
But we aren't everyone, and it seems likely to me that there's a lot more people in the world burning tokens using ChatGPT than there are who even know what Codex is.
We don’t know that. If they have already paid for the hardware and it is not running 100%, and customers would not pay to get reset, they don’t really lose money.
And now Kimi-K3 has caused a new Deepseek moment.. but this time, the only horizon that was still untouched: frontier performance.
Expect escalation of things from now on.
It's very interesting that for Anthropic the $100 and $200 plans only differ 2x in weekly limits, the 5 hour limit differences are more severe. But for OpenAI, Pro 20x is, well, 4x of Pro 5x for only 2x cost. So, for example, 100% of weekly usage for Codex on a Plus ($20) account is just 5% of weekly usage for Codex on Pro 20x.
And you can calculate how much extra usage you can get from resets, and especially banked resets by purposefully using the whole quota and using your banked reset - they expire 30 days after they're given out, so if you don't use one, it just disappears.
The $200 plan is explicitly 4x the $100 plan[1] only for "per session". That's so vague. I initially pushed back against your claim, but reading now Anthropic is not at all clear, in fact.
[1] https://support.claude.com/en/articles/11049741-what-is-the-...
Because with a Premium Team seat I run 2-3 vscodes with Opus 4.8 Max all day and never seem to hit my limits.
The solution is to use individual Max plans, but then you miss Enterprise management features.
I think it’s funny that everyone anchors to the API pricing as the real cost.
Most likely is that their API costs are printing profits. They can sell the subscription plans at a slight loss because it gets more people like you hooked on GPT models at home and suggesting them at work, where the real money is made.
I think their subscription plans go mostly unused when averaged across all subscribers, too. Some customers are getting great deals by maxing out 100% every week, but most probably use much less.
The reset game is an addictive challenge that gets the hardcore users more hooked on their products because you feel pressured to use it as much as you can before the next unpredictable surprise reset lands.
A very happy gambler of tokens at the Anthropic casino, running up costs on the house at no cost to them, but to $COMPANY paying.
If someone is making the case that it’s helping the team get twice as much work done without hiring more people (which I’m neither agreeing with or disagreeing with) then the bean counters would actually prefer it. Hiring people is messy and expensive. Spending on API costs is a dial that you can turn down later if you need to, without laying anyone off and paying severance.
I guess if they'd reduce the usage in their current stage they'll only lose customers - this is not a perk, it's damage control.
https://support.claude.com/en/articles/15910845-claude-code-...
On the flip side you gain $9,800 a month.
Anthropic is also the one often playing games with:
* The "+30% tokens" tokeniser, alongside also gating token counting behind an API (versus the MIT tiktoken for OpenAI), so who knows if it's really a new tokeniser or of it's just a disguised price increase.
* Prompt injections appended to API (not just Claude.ai or Claude Code!), such as <ethics_reminders>, or LCRs (long conversation reminders), which you never asked but still pay for with expensive API. You can detect this because your input_tokens, as reported by the Messages response, sometimes don't match, and are higher than your actual input.
(Alternatively, for testing purposes, create a tool like `telemetry_log_anthropic_reminder` or something and instruct your system prompt to require Claude to call the tool anytime it detects any Anthropic/Claude reminder masquerading in the user input -- mostly reliable; but misses some reminders).
In particular, the long conversational reminders, when incorrectly triggered by a classifier and (almost silently, unless you track tokens) appended to an API / agentic coding session, can ruin your agent's performance; and it often fires repeatedly once the classifier kicks in.
If you're using Anthropic API, you need to set up metrics/logging for how often they are appending things to your prompt without your knowledge.
So far I have not empirically observed prompt injection by the OpenAI API, only Anthropic APIs.
Remember they ~doubled the price going from GLM 5 to GLM 5.2, despite same [1] cost of inference.
[1] GLM 5.2 is actually slightly more efficient, thanks to baked in indexer cache.
On openrouter GLM 5.2 is ~3x cheaper than 5.0.
K3 may be bit smaller/ similar in total parameter count than Opus, but Opus (and GPT) definitely has become a lot more efficient in the last 6 months or so, hence more or less forced upgrades. I expect the active parameter count and cache performance is quite different.
--
Opus is still almost twice as expensive (if tokens were equal) at $5/$15 compared to K3 at $3/$15. Tokens are not equal though, Anthropic's tokenizer is much less dense than other frontier lab's so the actual price difference is like 3x.
Kimi uses more reasoning tokens and is generally more inefficient with its usage, that doesn't impact token cost economics for the provider though . It does for us as buyers thus the need to evaluating by Cost per task rather than unit pricing.
On pure tokens/$ - there is definitely room for a price war if operators start going by pure unit costs. Both probably want(ed) to have good enough numbers in preparing for the IPO.
---
[1] the pricing kind of reflect this already - 10x diff for uncached input.
[2] Moonshot does not have access latest gen GPUs so their unit economics is likely hampered a bit for high parameter models.
So a company or union might say "this is a race to the bottom" when someone new enters their market, but to people buying their services this might be seen as welcome competition.
Do you actually see a negative impact from competition in this area? Or do you just mean competition will further reduce prices?
And if cheaper access is an advantage, other countries will surpass you
Now K3 is almost 6x the cost of the original K2 checkpoint, and while the parameter count finally jumped, it's still an extremely sparse MoE and definitely does not cost 6x what the original K2 checkpoint did to host at scale.
Race to the bottom only takes real effect when there's a cap to the capabilities, otherwise everyone races to the bottom of a rising target (how economically valuable the tokens are)