Muhammad Yusuf Rakhimov

Tech Entrepreneur | Founder and CEO at OnCruit.

Documenting my journey | Vibe Engineering vertical startups across multiple industries

[Muhammad Yusuf Rakhimov @oncruit]

some thoughts on Fable 5 / Mythos-class models:
i think this is one of the more important AI releases, but not only because of benchmark numbers.
the interesting part is the shape of the release.

Anthropic basically split the model into two ideas:
Fable 5: the public, safeguarded version
Mythos 5: the more restricted version for trusted access
that framing matters.
it says we’re entering a phase where the most capable models may not be released as one clean public product anymore.
instead, they’ll come with tiers of access, domain-specific guardrails, fallback systems, usage limits, and trusted-access programs.
that is probably the future of frontier AI.
not just “new model dropped.”
more like:
new capability class dropped, but the release surface is controlled.
Fable 5 feels like Anthropic trying to bring Mythos-level capability to normal users without fully opening the dangerous parts of the system.
and honestly, this is probably the correct direction.

because the whole point of a Mythos-class model is not just that it writes better code or answers harder questions.
the point is longer-horizon agency.
models are getting better at taking a messy objective, holding context, making plans, using tools, dealing with ambiguity, and pushing work forward without constant babysitting.
that is the real unlock.
chat was the first era.
coding agents were the second era.
long-horizon autonomous work is the third era.
and Fable 5 is interesting because it feels designed for that third era.
not “answer my question.”
more like:
understand the objective
manage the context
make progress across time
handle messy work
come back with something useful
that is a different product category.

but the brutal part is cost.
Fable 5 is not cheap.
at $10 per million input tokens and $50 per million output tokens, this is not the model you throw at every tiny task just because it is shiny.
that would be dumb.
this is where people need to stop thinking like model fanboys and start thinking like operators.
Fable 5 should not replace every model in your stack.
it should sit at the top of the stack.
use it where failure is expensive.
hard planning
complex coding
agent orchestration
research synthesis
messy business reasoning
high-stakes workflows
long tasks that require judgment
for everything else, use cheaper models.

this is the same pattern that is becoming obvious across the whole AI market:
premium models for judgment
cheap models for scale
specialized models for modality
routing layer in the middle
the winning companies will not be the ones saying “we use Fable for everything.”
the winning companies will be the ones that know exactly when Fable is worth paying for.
that is the real game now.
cost per successful task.
not benchmark score.
not vibes.
not leaderboard screenshots.
cost per successful task.
if Fable 5 can solve a task in 20 minutes that would take a human 6 hours, the price is irrelevant.
if it is being used to summarize basic CRM notes, the price is stupid.
same model.
different economics.

this is why i think model routing is becoming a serious product discipline.
the question is no longer:

“which model is best?”
the question is:
“which model should handle this specific task, at this risk level, with this context size, at this acceptable cost?”
that’s a much more mature way to think.

the Mythos/Fable split also shows something else:
safety is becoming part of product architecture.
not a side note.
not just a policy page.
actual routing, restrictions, fallback behavior, access levels, and domain boundaries.
some people will hate that.
some people will call it over-controlled.
but for frontier models that can operate across cybersecurity, biology, chemistry, software systems, and enterprise workflows, this was probably inevitable.
the more capable the model gets, the less likely it is that everyone gets the same unlimited version.

so my current read:
Fable 5 is not just a better Claude.
it is a signal.
a signal that frontier AI is moving from chatbots to controlled capability systems.
the model is only one part of the product.
the harness, routing, safeguards, access layer, pricing, and deployment surface are becoming just as important.
that is where the next wave of AI products will be built.
not around one magic model.
around knowing how to use different levels of intelligence correctly.

Fable 5 is expensive.
but if it really extends the time horizon of useful autonomous work, it might still be cheap.
that’s the paradox of frontier AI now.
the best models will look overpriced for small tasks and underpriced for serious work.
the skill is knowing the difference.

[Muhammad Yusuf Rakhimov @oncruit]

@GergelyOrosz @mattpocockuk I want live process of his vibecoding using the skills

[Muhammad Yusuf Rakhimov @oncruit]

@Gorden_Sun It was like a a year ago that this feature was introduced

[Muhammad Yusuf Rakhimov @oncruit]

@ajaypatel_aj @shadcn How are you guys using shadcn ui on apps or crm? I vibecoded a crm on my pc and prompted to use shadcn style ui but it doesnt look good

[Pavel Durov @durov]

Telegram has applied for the .gram domain zone.

If the application is approved by ICANN, a billion Telegram users could get their own second-level domains — yourname.gram.

Users would be able to set up their interactive websites hosted by Telegram — with one prompt ✨

[Muhammad Yusuf Rakhimov @oncruit]

@Motion_Viz Stack

[Muhammad Yusuf Rakhimov @oncruit]

There will not be an option to buy usage reset. Trust me

[Muhammad Yusuf Rakhimov @oncruit]

@ibocodes Migration took me 3 hours with my codex

[Muhammad Yusuf Rakhimov @oncruit]

@itsivanfalco Pack

[Muhammad Yusuf Rakhimov @oncruit]

going hard as soon as I read this tweet from Tibo

[Tibo @thsottiaux]

Old news actually from a bunch of days ago, but crossed that 15M. Enjoy a nice reset everyone. Landing in the next hour or so, go /fast.

[Muhammad Yusuf Rakhimov @oncruit]

I forgot to press my usage reset ‘button’ when I had only 30% left

[Muhammad Yusuf Rakhimov @oncruit]

Shipping day

[Muhammad Yusuf Rakhimov @oncruit]

does anybody know what does "optimizing the conversation mean" in here?

[Muhammad Yusuf Rakhimov @oncruit]

@pcshipp Cloudflare

[Muhammad Yusuf Rakhimov @oncruit]

@theo but 6 hours prior to this reset, it started bleeding usage rate limits))

[Muhammad Yusuf Rakhimov @oncruit]

Tried @mattpocockuk's Improve Codebase skill on RecruitOS.

It inspected the repo and generated a visual architecture review showing where responsibilities were leaking between modules.

One example: our driver readiness/compliance interpretation was spread across Today, Talent, dossier and clearance flows.

The skill proposed one deeper module to own that logic instead.

It even found a 1,244-line candidate dossier implementation with 22 reads + 8 history sources and mapped out a cleaner architecture.

Really interesting way to use coding agents beyond just generating features.

[Muhammad Yusuf Rakhimov @oncruit]

My codex usage limit started bleeding for the last 4 hours, never had that before

[Muhammad Yusuf Rakhimov @oncruit]

@carrierwhisper1 https://t.co/BYLXpVy86l

[Muhammad Yusuf Rakhimov @oncruit]

@RobIW_dev @MarcosHernanz how are you guys using so many subagents all at once?

[Muhammad Yusuf Rakhimov @oncruit]

@AmirMushich Wheres prompt

[Muhammad Yusuf Rakhimov @oncruit]

Nasib qilsin

[Noah Tran @_nghiatran]

Ordered €7500 M5 Max as a gift for my hard work 😊