You were lied to about Fable
TL;DR
The viral benchmark is unreliable: It ranks Claude Sonnet 5 as the best reasoning model ever and puts GLM 5.2, Grock 43, and Neotron 3 Ultra above Opus 4.8, which immediately discredits its claims about Fable 5 coding performance.
Anthropic's safety system uses a two-stage classifier: A cheap probe monitors Claude's internal activations first, and only triggers an expensive classifier if sensitive regions light up, explaining why cryptography puzzles trigger refusals even when they are harmless.
Fable leaving subscriptions on July 7th is a capacity experiment, not a permanent price hike: Anthropic needs usage data from power users to calculate GPU allocation, and Thoric confirmed they aim to restore it to subscriptions as soon as capacity allows.
Max and x-high effort settings are a waste of money: The speaker has never seen max produce better results than high, but it costs 10 to 50 times more tokens.
Routing sub-agents to cheaper models slashes costs: The speaker taught Claude Code to use Codex for token-heavy tasks like PDF processing and computer use, letting Fable orchestrate rather than grind through every task itself.
The speaker merged 15+ pull requests for roughly $150-200 in token costs: He estimates he got a month of work done in three days using the model.
Summary
Viral benchmarks claiming Fable 5 got dumber at coding come from a source that also ranks Claude Sonnet 5 above Opus 4.8 at reasoning, which should tell you everything.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
The Retirement Email Isn't a Warning
Model retirements now arrive every few weeks; the config-eval-rehearsal loop turns each deprecation email from a fire drill into an afternoon swap.
Playbook
The Cheapest Model That Passes
OpenRouter lists 400 models behind one API. The fix for choosing isn't a better leaderboard, it's a four-step protocol that ends in a real eval.
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.