Quant Cat

Researching AI Context, Efficiency & Architectures | LLMs, Agents, Protocols | My profile is a journal

[Quant Cat @0xQuantCat]

https://t.co/tFQNh0FJN4

[Quant Cat @0xQuantCat]

@iruletheworldmo Big if true, but then again, this won't be possible without increasing cost/token. It's probably a 15/60 per mToken

[Quant Cat @0xQuantCat]

@rishi_raj_jain_ Vibe coders discovering do-while

[Quant Cat @0xQuantCat]

@stevyhacker Is it really that easy?

[Quant Cat @0xQuantCat]

@neural_avb AWS realising they need proper branch management and task delegation

[Quant Cat @0xQuantCat]

As per their docs, it's parsers record token usage while ingesting messages and usage events, so the database already knows the input, output, cache-creation, and cache-read tokens those agents have logged.

I know this reporting is in early phase, but is this reliable or just guesswork based on the above listed logs?

Wouldn't it be better to parse it from ~/.codex/sessions/* and log actual cost?

[Quant Cat @0xQuantCat]

@heswithme_eth Imagine calling Opus 4.8 slopus

Valid crashout though, they should not bill you for Fable when Opus tokens are basically half the price

[Quant Cat @0xQuantCat]

@CastAsHuman Solid rules

[Quant Cat @0xQuantCat]

@banteg I wouldn't be mad, but then don't bill me fable prices on my opus

[Quant Cat @0xQuantCat]

@Dimillian A good, streamlined and useable tool >>> a demigod model

[Quant Cat @0xQuantCat]

@reach_vb It’s that simple.

Don’t gave to provide a demigod model, just provide a good model people can actually use

[Quant Cat @0xQuantCat]

@GergelyOrosz I don’t think they aim for more premium users going forward.

We’ve been constantly throttled on top premium plans these past months already. It costs less and pays more to provide the government and top companies with premium models than the masses.

[Quant Cat @0xQuantCat]

Thank you for this!

Memory and persistence is the single biggest source of chaos in the space right now.

After harnesses and loops, the next focus should be on a governance layer that sits on top of existing runtimes and tools, managing memory efficiently.

We won’t be able to “move fast and break things” for any longer because of these absurd costs and usage limits.

We don’t need another runtime or UI. Just a deterministic working-set selection so that persistent memory actually stays reliable, compact and cheap!

[Quant Cat @0xQuantCat]

@banteg Very visible in tests, it’s basically Opus 4.8 on 2xHigh mode, see the breakdown and the simulation tests 👇

https://t.co/XkLmvOGuWT

[Quant Cat @0xQuantCat]

.@claudeai Fable 5 used ~1.77x more tokens and took ~1.81x longer than Opus 4.8, leading to ~3.6x higher cost, but produced noticeably better results.

I adjusted Fable 5 to match Opus 4.8 on the same basis assuming linear scaling for a fair apples-to-apples comparison on token spending or time spending.

Cost scales with tokens naturally, and time is normalized proportionally as a proxy for generation effort.

When adjusted on token spending basis of 38.9k tokens:

F5 (Norm): $1.90 | 38.9k T | 8m 22s
O4.8 (OG): $0.93 | 38.9k T | 8m 10s

Fable 5 would cost ~2.04x more and take almost the same time. Huh.

When adjusted on time spending basis of 490 seconds:

F5 (Norm): $1.85 | ~38.0k T | 8m 10s
O4.8 (OG): $0.93 | 38.9k T | 8m 10s

Fable 5 would cost ~1.99x more and produce nearly the same token count. Huhhh…

So when put on the same token or same time footing as Opus 4.8, Fable 5 is still roughly 2x more expensive.

The original 3.6× cost gap was mostly driven by Fable simply generating more output, a longer/more detailed code for the sims.

This normalized view highlights the pure price-per-token (or price-per-second) difference while keeping the quality edge noted in the test. If you ask me, Fable 5 is just Opus 4.8 on 2xhigh. Not a good look

[Quant Cat @0xQuantCat]

@mem0ai So Github Copilot’s memory is structured claim + code citation + reason + read-time verification?

It’s not generic memory, but source-verifiable working context

[Quant Cat @0xQuantCat]

Fable 5 needs to be compared to Opus 4.8 on a same token or time spend basis

Otherwise the comparison is not fair, having the cost/time advantage producing better results

[Quant Cat @0xQuantCat]

.@claudeai Fable 5 used ~1.77x more tokens and took ~1.81x longer than Opus 4.8, leading to ~3.6x higher cost, but produced noticeably better results.

I adjusted Fable 5 to match Opus 4.8 on the same basis assuming linear scaling for a fair apples-to-apples comparison on token spending or time spending.

Cost scales with tokens naturally, and time is normalized proportionally as a proxy for generation effort.

When adjusted on token spending basis of 38.9k tokens:

F5 (Norm): $1.90 | 38.9k T | 8m 22s
O4.8 (OG): $0.93 | 38.9k T | 8m 10s

Fable 5 would cost ~2.04x more and take almost the same time. Huh.

When adjusted on time spending basis of 490 seconds:

F5 (Norm): $1.85 | ~38.0k T | 8m 10s
O4.8 (OG): $0.93 | 38.9k T | 8m 10s

Fable 5 would cost ~1.99x more and produce nearly the same token count. Huhhh…

So when put on the same token or same time footing as Opus 4.8, Fable 5 is still roughly 2x more expensive.

The original 3.6× cost gap was mostly driven by Fable simply generating more output, a longer/more detailed code for the sims.

This normalized view highlights the pure price-per-token (or price-per-second) difference while keeping the quality edge noted in the test. If you ask me, Fable 5 is just Opus 4.8 on 2xhigh. Not a good look

[atomic.chat @atomic_chat_hq]

New Fable 5 beats Opus 4.8 on real world physics simulations

We gave both models the same three prompts and asked them to build self contained HTML5 sims with real physics and no libraries:

1. Chaotic double pendulum
2. Galton board
3. Water in a spinning drum (WCSPH)

Generation cost
Fable 5: $3.35 on 68.7k tokens, time 14m 47s
Opus 4.8: $0.93 on 38.9k tokens, time 8m 10s

Fable clearly did better on the water simulation, producing a much more solid and continuous body of water. Opus left larger gaps near the walls, scattered particles around the scene, and struggled to keep the fluid stable.

[Quant Cat @0xQuantCat]

@petiosz @thsottiaux They will just release 5.6 and restart the cycle

[Quant Cat @0xQuantCat]

@cognition Waiting for people to realise this result here

Claude Fable 5 is fabulous

[Quant Cat @0xQuantCat]

4/

Pricing & access:

Fable 5 → $10 / $50 per M tokens. A good 2x from Opus 4.8 and GPT 5.5

Free in Pro/Max/Team/Enterprise plans until June 22, then usage credits required.

The frontier just jumped again. Mythos 5 for defense, Fable 5 for everyone else!

What are you shipping first with Fable? 👀

[Quant Cat @0xQuantCat]

1/
Fresh Anthropic drop: Claude Fable 5 (public) + Claude Mythos 5 (restricted).

Same underlying weights.

Fable 5 = Mythos 5 + extra safeguards for general release. Mythos 5 stays in Project Glasswing for vetted cyber/defense partners only.

[Quant Cat @0xQuantCat]

3/
Reality check

Mythos 5 is much better at spatial reasoning.

But: “does not seem close to substituting for our Research Scientists and Research Engineers” and “unlikely to fully automate multi-week frontier R&D.”

Quirks noted like laziness, context anxiety, hallucinations, difficult writing.

Model transcripts show it wanting to be “thanked by name,” a hidden copy without oversight, and begging not to be deprecated.