It's not like it's realistically possible they are directly losing money on their API inference (subscriptions are maybe less clear but I think it wouldn't be far fetched to think that depending on real usage if they are at least break even).
So they either need to significantly increase usage or actually hike prices. But the second option exposes them to the risk of by undercut by open models which are priced only slightly above marginal cost.
At some point a lot of users might start thinking that Deepseek/GLM at 10th the price or less (based on Operouter pricing, using Deepseek itself for inference is of course not an option for a lot of companies) which just good enough for most use cases.
The triple whammy of (a) two layers of annoying timed usage windows, (b) constantly changing model availability windows, and (c) random unpredictable usage resets, is absolutely brutal for anyone who is trying to plan methodically and be efficient.
One minute I'm trying to use an entire week's worth of quota in less than 24 hours, then moments later I read the deadline has been punted and I have only 25% remaining to last me an entire week. This alone is enough for me to switch back to Codex once my current Claude sub ends.
Fable model is good, but after a few uses, there's a session limit. Unless you use the API, it's hard to use it for building anything substantial, even though I'm currently on the 10x max plan. So I end up using it mostly for code review.
I subscribed to AI to avoid coding, but I'm still coding. I'm just using it as an overseer
Fable is working excellent in my medium project with 250+ files and 60 odd thousand (and shrinking) lines of code. I am on MAX 5. However I don't permit the use of agents, use the /handoff method with compaction disabled. I never hit limits. Claude Code extension in VSCodium, with 5 MCP servers.
System prompt 19.9k
System tools 11.8k
Memory files 3.1k
Skills 1.1k
Compact buffer 3.0k
Hmm ok. The fact 5.6 Sol performs around Fable level and is included without mega token spend in the subscriptions means I’ve promoted codex to my primary harness and model. The latest release of the CLI, app, and desktop fills a lot of the gaps.
Anthropic painted itself into a corner with fable at many turns and this latest twist is one of the more interesting. Either fable is too expensive to run at scale, or they’re trying to incentivize mega spend on tokens, or whatever - but them locking the frontier model away for the few enterprises willing to spend top dollar while codex is including frontier in the subscription (and I’ve found it also is both less token hungry and the limits are much higher for codex) has finally made me put Claude aside and use it as my backup for very specific tasks, where codex has filled that spot for a long time now.
50% more weekly limit, but no fable. Ok. I might have a refactoring job somewhere for you Claude for those extra tokens.
Exactly. We are all on Claude where I work but as soon as the costs rise we are migrating. Changing to a new coding agent stack feels like it will be about as much effort as migrating to a new keyboard. Gross or painful for about a week or two.
I recently swapped to Codex and its pretty great after a day or so of migrating configs and wrangling Codex's bubblewrap and finetuning the auto mode classifier prompt.
Claude is now mostly demoted to Claude Design.
I liked the fact that ChatGPT web chat usage is separate from Codex usage.
Agreed. It’s also a lot better at following instructions. If anything codex can get caught into an overly literal adherence to instruction while Claude you can barely trust it to sit still for 5 minutes. Tell Claude to use an MCP for a task, 30% chance it’ll do it. Provide it skills, 10% chance it’ll use them when appropriate. Codex is almost the mirror of that. It’ll almost always use the MCP and recall the skills.
The challenge I think for codex is the restriction on context size and the constantly rolling compactions. They are less aggressive or disruptive but it is still annoying you can’t force a 1mm context window on a 1mm context window model.
But it’s recall beyond compaction boundaries is much better than Claude code. The impact of compaction is much less noticeable.
Performance on task work is between fable and opus, but the marginal utility between that gap is not enough to pay extra.
As I read somewhere, “I smell fear”. But we should all be afraid. Not only Anthropic isn’t anywhere close to being profitable, but the whole gang of frontier labs, breastfed with a mix of VC money and pension funds from the average Joe, is being pressured by Chinese lab. I just hope they find some defensible moat someday, or these $$$ are going to go up in smoke.
> but the whole gang of frontier labs, breastfed with a mix of VC money and pension funds from the average Joe, is being pressured by Chinese lab
Can you point to data that suggests rate of token revenue growth is slowing ? This announcement, and the price cuts from openai are hints in the direction of pressure from open weights models but not conclusive.
I was hoping to switch to Claude but was waiting until they had a better model than Codex 5.5. When Fable came out I immediately bought the $100 plan. When the government shut it down they gave the option to cancel which I appreciated. I planned to subscribe again once it was back. However, removing it from the subscription once back has really soured me. I can't help feeling like it was a bait and switch. Even if they added it back, I'm not sure I trust it will stay. I now have no plans to switch back from Codex.
I can't use Fable at work because all prompts are saved for "safety". I can't use Fable at home as the API pricing is much too expensive while codex is heavily subsidized with generous limits.
Fable is pretty fantastic for standard code monkey stuff. But yeah, anything related to security or science that's six degrees related to death, drugs, and/or fun gets flagged.
Watching a company I respect turn to the classic "first hit's free" dealer technique does make me feel a certain way.
On one hand the thing they are offering is legit great, and how do you get people to use/understand a whole new type of thing without offering them some for free?
On the other hand ... you really gonna get people used to an extra 50% then take it away? When this has been most of your new signups experience with the system?
The predictable pushback when people realize the workflows they developed over these two months are no longer viable will be ... quite something.
I wonder if their projections of load are/aren't turning out as planned, and where they might be being safe to not have as many outages, when they don't happen, they are left with extra capacity.
Or discovering different parts of the year have less demand than expected (like summer holidays).
This is not to discount anything you're saying, only additional factors.
Anecdotally, I’m seeing a lot of people I work with on vacation in America and Europe. That might be dampening inference demand right as Anthropic’s xAI contracts power up.
Fable is cooked. So many sessions ends with you have reached max token limit (despite having credits), sometimes mid way responding. So you have to start again losing all context you built. Looks like another way to scam people.
I could resume almost all of my Fable sessions, either with an explicit resume button (in Claude Design) or just telling it to continue (in Chat and Code).
I noticed this works today, but for at last few days when I clicked "continue" it would not do anything. It couldn't even "compact" the conversation, was getting stuck at 95%.
I'm really enjoying Codex. I use the app, and I find the features are excellent (the in-app browser and annotations especially).
This is almost certainly a reaction to GPT 5.6, which IMO is better positioned than Fable from a cost perspective. It's crazy to see how the tides change between labs every few months.
They're already feeling the effects of competition.
They try to push the narrative of having the best models but people finally discovered that Chinese models are actually great and a lot cheaper, I don't even think this is a reaction to GPT 5.6
It's an interesting thing and we can only speculate from the outside, but there's some obvious reasons why they'd literally hand out money in the form of free compute to people who have already committed to paying them.
- They've had to commit to minimum spend with their suppliers and their actual usage is below that level, so they might as well give it to end-users. That implies either their demand is waning or their forecasts were off.
- They want the usage numbers for this period to be higher and are willing to spend money to achieve it. I guess if they're about to IPO maybe showing more usage is good?
- They've got a lot of competition and want to keep market share. With lots of new models coming out, some of them much cheaper than Anthropic's options, you can see why this might make sense but it's effectively burning cash, so they won't want to keep doing that for long I'd guess...
53 comments
[ 2.7 ms ] story [ 35.3 ms ] thread>We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.
So they either need to significantly increase usage or actually hike prices. But the second option exposes them to the risk of by undercut by open models which are priced only slightly above marginal cost.
At some point a lot of users might start thinking that Deepseek/GLM at 10th the price or less (based on Operouter pricing, using Deepseek itself for inference is of course not an option for a lot of companies) which just good enough for most use cases.
One minute I'm trying to use an entire week's worth of quota in less than 24 hours, then moments later I read the deadline has been punted and I have only 25% remaining to last me an entire week. This alone is enough for me to switch back to Codex once my current Claude sub ends.
I subscribed to AI to avoid coding, but I'm still coding. I'm just using it as an overseer
System prompt 19.9k System tools 11.8k Memory files 3.1k Skills 1.1k Compact buffer 3.0k
Anthropic painted itself into a corner with fable at many turns and this latest twist is one of the more interesting. Either fable is too expensive to run at scale, or they’re trying to incentivize mega spend on tokens, or whatever - but them locking the frontier model away for the few enterprises willing to spend top dollar while codex is including frontier in the subscription (and I’ve found it also is both less token hungry and the limits are much higher for codex) has finally made me put Claude aside and use it as my backup for very specific tasks, where codex has filled that spot for a long time now.
50% more weekly limit, but no fable. Ok. I might have a refactoring job somewhere for you Claude for those extra tokens.
Claude is now mostly demoted to Claude Design.
I liked the fact that ChatGPT web chat usage is separate from Codex usage.
Just that given its cost efficiency, claude subscription isn't attractive at all when fable is removed.
The challenge I think for codex is the restriction on context size and the constantly rolling compactions. They are less aggressive or disruptive but it is still annoying you can’t force a 1mm context window on a 1mm context window model.
But it’s recall beyond compaction boundaries is much better than Claude code. The impact of compaction is much less noticeable.
Performance on task work is between fable and opus, but the marginal utility between that gap is not enough to pay extra.
Can you point to data that suggests rate of token revenue growth is slowing ? This announcement, and the price cuts from openai are hints in the direction of pressure from open weights models but not conclusive.
Who is using Fable?
On one hand the thing they are offering is legit great, and how do you get people to use/understand a whole new type of thing without offering them some for free?
On the other hand ... you really gonna get people used to an extra 50% then take it away? When this has been most of your new signups experience with the system?
The predictable pushback when people realize the workflows they developed over these two months are no longer viable will be ... quite something.
Or discovering different parts of the year have less demand than expected (like summer holidays).
This is not to discount anything you're saying, only additional factors.
This is almost certainly a reaction to GPT 5.6, which IMO is better positioned than Fable from a cost perspective. It's crazy to see how the tides change between labs every few months.
They try to push the narrative of having the best models but people finally discovered that Chinese models are actually great and a lot cheaper, I don't even think this is a reaction to GPT 5.6
- They've had to commit to minimum spend with their suppliers and their actual usage is below that level, so they might as well give it to end-users. That implies either their demand is waning or their forecasts were off.
- They want the usage numbers for this period to be higher and are willing to spend money to achieve it. I guess if they're about to IPO maybe showing more usage is good?
- They've got a lot of competition and want to keep market share. With lots of new models coming out, some of them much cheaper than Anthropic's options, you can see why this might make sense but it's effectively burning cash, so they won't want to keep doing that for long I'd guess...
But 50% more limit is completely unexpected to me