When you build something people REALLY want, you can pull of ridiculous stunts like Claude's pricing:
- Vague limits
- Weekly/monthly usage stats ??
- 50% of your usage on Fable 5
- $200/mo subscription that doesn't include Fable 5
- But you can use it with "usage credits"
I'm frustrated but am also hooked. Kinda of a lesson in startups in a way.
Fable isn't needed except deep RE work. Opus 4.8 xhigh can code anything you throw at it and can code to public interfaces extremely well. I have rarely seen it make code level mistakes. In fact its the only model I feel that can actually code and I can actually trust to code. That's imo quite valuable.
Claude Pro with Sonnet 5 at $20 per month for a limit I never reach is OK. I'm using it 3 days per week full time.
I'm also experimenting with DeepSeek 4 and it seems on par. Its API price is much cheaper and it's top up, not a subscription. I still didn't consume the $2 I loaded on my account sometimes in June. However they about to charge double price in what seems to be China's morning and afternoon.
are you using it to write code? $20/mo was never enough and I'm not even full time on coding .. yeah different models are worth experimenting .. but I'm okay with paying higher for the best model .. just need predictability and stability of pricing, which Claude doesn't give, and yet we still use and pay for .. because it's such a good product.
Switching to Codex if their subscription includes 5.6 and Anthropic’s doesn’t include even some small quota of Fable. I’d be using local LLM’s if I didn’t want any access to frontier
Yep, that's me. the only real blocker is that American companies don't trust Chinese providers, but i could just find a good American provider that hosts DeepSeek and/or GLM. I would at least be able to choose my own agent instead of a quite mediocre one that wastes time and output nonsense verbs in a pathetic attempt to gain sympathy.
The only reason that stopped me from doing it is the absence of a subscription, and I did believe I couldn't get the same value with API pricing, but I'm starting to see that it's a blatant lie and true only for anthropic and openai...
I'm also experimenting with DeepSeek 4 and it seems on par. Its API price is much cheaper and it's top up, not a subscription. I still didn't consume the $2 I loaded on my account sometimes in June. However they about to charge double price in what seems to be China's morning and afternoon.
Claude Code is good, but a lot of the cost is repo discovery: finding files, reading candidates, following imports, etc.
Fabric indexes the repo locally and gives Claude Code the relevant context up front. Nothing goes to HailMary.
Tiny early test on httpx: 58% lower API cost across 2 tasks, both passed tests.
Repo is here if anyone wants to try it or poke holes in it: https://github.com/hailmarylabs/fabric/
I doubt it.
It builds bad will, that will have users say "fuck them" on the first chance they get.