Two frontier models, two career stages: one at the peak of its game, the other at the very beginning.

The Veteran

GPT 5.6

Peak of its game — the best, and likely the last, of the GPT-5 training run.

The Rookie

CLAUDE FABLE

A monster just entering the league — raw talent, barely post-trained.

The Career Curve

GPT-5 series Fable's new class

Illustrative trajectories — Berman’s characterization of the two models, not benchmark data.

The Tape

Likely the last model of the GPT-5 training run — the very best of that class

Status

First version of a new class — “an absolute monster,” just entering the league

Efficient — “finds very direct paths to solutions”

Game

“Technically the smartest model on the planet”

“Just reliable. It has good game IQ.”

Temperament

Raw talent — the polish isn’t there yet

At the peak — little room left to grow the GPT-5 series

Ceiling

Post-training has only just begun — “so much room to grow”

Generous quotas & resets, backed by an early all-in bet on GPUs

Supply

Big and expensive to serve — Anthropic is squeezed on capacity

In His Words

“It’s incredibly smart. It gets things done well. It’s efficient… It’s just reliable. It has good game IQ. It is the vet.” — ON GPT 5.6 · 12:31

“Fable is this absolute monster of a model, but it’s like a rookie — it’s just entering the league… that raw talent is there.” — ON CLAUDE FABLE · 12:42

What Happens Next

The Sequel

GPT-6 is likely already baked — OpenAI is probably deep in post-training now, prepping broad release. The GPT-5 series itself is near done.

The Serving Problem

Fable-class models are bigger and costlier to serve. OpenAI’s GPU overinvestment is its hedge; Anthropic is already living the crunch.

The Wildcard

Recursive self-improvement. “Once RSI starts, whoever’s in the lead wins” — the bet Anthropic may be making.

THE CALL

Budget for only one subscription? Berman recommends OpenAI — on value, quotas, and predictability — “although Fable is slightly better right now.” And either way: use your quotas. Build.