I've not run into the same Fable quota problems that others have reported myself yet. I wonder if people are running prompts that are somehow spinning up a ton of subagents?
Sometimes I think about how much power I'm using with 10 Fable agents running in parallel for 10 minutes each. It's gotta be a lot. Compared to my PC which with a great GPU can run a crap agent that takes way too long and does a terrible job at everything.
Anyways, I don't get it, do you wish they never released Fable? I can't believe they're making any money on it. I'd bet they said we're gonna burn 100 million dollars on 12 days of unlimited access just to at least give people a taste and maybe recoup our losses if people actually pay for it later.
Currently running an experiment where I'm using Fable as a planner/orchestrator/verifier and using Mimo 2.5 as subagents/builders. Seems to be doing well with my old specs.
Too expensive, yet probably still too cheap to make up for their cost of developing and running it. Don’t expect prices to go down. Or if they superficially go down, the models just consume more tokens per task so your bill still increases in the end.
It's only available on the $100-$200/month subscription plans until June 22nd, after which they're going to be charging everyone the full API token price. Then it's going to be really expensive.
(I think it may even be expensive enough for them to recoup their costs, unless OpenAI or Gemini put out a similar capability model before Anthropic have had a few months to make bank from it. There was that rumored customer who spent $500m in a single month, and that was just for Opus!)
Ah, well that makes sense. Will be interesting to see what adoption looks like for a model that’s billed at a more realistic price-point towards profitability.
Currently Fable is available through Chinese resellers at around ~$17.50/1M tokens, or about a 65% discount vs API prices. Still expensive compared to Opus.
Fable so far has only been incrementally better than Opus 4.8 in my testing. The testing so far has been coding on a large monolith. In general Claude is starting to feel like it's optimize for greenfield vibe-coding projects. I vastly prefer Codex for coding on large, established codebases.
Well, as predicted I see many here complaining about how expensive Fable is. [0]
It is not just companies that know it is economically expensive, but it is also expensive to companies who have zero-data retention policies and they cannot offer AI features that use this model.
Anthropic doesn't care and won't talk to you unless you're paying them hundreds of millions. There are now local models which are great alternatives which have no quotas and can work with companies that have zero data retention policies.
None of those models are Fable's class. There are definitely many cheaper options for LLM agents, but people are paying attention to Fable for a reason.
People are also paying attention to open models for a reason, they are around 6-9 months behind and far cheaper. Fable output looks like a modest improvement, certainly not 2x like the price. We'll have similar models we can run ourselves by this time next year
Suggestion: run this command:
That should tell me the token total and dollar cost across all of your local coding agents.If it doesn't seem to know how much Fable costs you can follow these instructions to teach it the Fable pricing scheme, then run the command a second time to see the results: https://til.simonwillison.net/llms/agentsview-custom-model-p...
I've not run into the same Fable quota problems that others have reported myself yet. I wonder if people are running prompts that are somehow spinning up a ton of subagents?
Sometimes I think about how much power I'm using with 10 Fable agents running in parallel for 10 minutes each. It's gotta be a lot. Compared to my PC which with a great GPU can run a crap agent that takes way too long and does a terrible job at everything.
Anyways, I don't get it, do you wish they never released Fable? I can't believe they're making any money on it. I'd bet they said we're gonna burn 100 million dollars on 12 days of unlimited access just to at least give people a taste and maybe recoup our losses if people actually pay for it later.
Currently running an experiment where I'm using Fable as a planner/orchestrator/verifier and using Mimo 2.5 as subagents/builders. Seems to be doing well with my old specs.
How do you achieve that? Any special tooling or just skills?
Too expensive, yet probably still too cheap to make up for their cost of developing and running it. Don’t expect prices to go down. Or if they superficially go down, the models just consume more tokens per task so your bill still increases in the end.
It's only available on the $100-$200/month subscription plans until June 22nd, after which they're going to be charging everyone the full API token price. Then it's going to be really expensive.
(I think it may even be expensive enough for them to recoup their costs, unless OpenAI or Gemini put out a similar capability model before Anthropic have had a few months to make bank from it. There was that rumored customer who spent $500m in a single month, and that was just for Opus!)
Ah, well that makes sense. Will be interesting to see what adoption looks like for a model that’s billed at a more realistic price-point towards profitability.
Currently Fable is available through Chinese resellers at around ~$17.50/1M tokens, or about a 65% discount vs API prices. Still expensive compared to Opus.
... how. how is that even possible. pirated usage plans?
They're reselling pooled capacity from many Max accounts through Chinese proxies at below official prices:
https://www.chinatalk.media/p/the-grey-market-for-american-l...
Account sharing. They route your request via one of thousands of Max 5X accounts.
Fable so far has only been incrementally better than Opus 4.8 in my testing. The testing so far has been coding on a large monolith. In general Claude is starting to feel like it's optimize for greenfield vibe-coding projects. I vastly prefer Codex for coding on large, established codebases.
Guys, are we poor? I just ran out after a 20 minute refactoring exercise for the next 4.5 hours lol.
built this to stop fable blowing up my usage limit https://github.com/teamchong/pxpipe
Well, as predicted I see many here complaining about how expensive Fable is. [0]
It is not just companies that know it is economically expensive, but it is also expensive to companies who have zero-data retention policies and they cannot offer AI features that use this model.
Anthropic doesn't care and won't talk to you unless you're paying them hundreds of millions. There are now local models which are great alternatives which have no quotas and can work with companies that have zero data retention policies.
[0] https://news.ycombinator.com/item?id=48467112
OpenCode has a generous $10 plan for open weight models, I'm quite happy with it
https://opencode.ai/docs/go/
Fireworks.ai has a $50 plan in early access
https://docs.fireworks.ai/firepass
I'm shopping around for more good deals like this, I'm long on open weights and sour on Big Ai
None of those models are Fable's class. There are definitely many cheaper options for LLM agents, but people are paying attention to Fable for a reason.
People are also paying attention to open models for a reason, they are around 6-9 months behind and far cheaper. Fable output looks like a modest improvement, certainly not 2x like the price. We'll have similar models we can run ourselves by this time next year
Wait till June 22nd! That'll be like $5000/week usage. This may be the cheapest Fable-class or better Anthropic model you ever see in your lifetime.