Anthropic's Fable 5 returns after a 19-day export control suspension with enhanced safeguards, while OpenAI slashes inference costs by 50% for unauthenticated users using undisclosed optimization. The episode also covers AWS's $1 billion investment in forward-deployed AI engineering, Base44's launch of a narrowly fine-tuned coding model, and debates on using Fable 5 for strategic thinking rather than routine coding.
Summarized by Podsumo
Anthropic's Fable 5 returns after 19 days with enhanced guardrails (99% effective classifier) but may fall back to Opus 4.8 on routine coding tasks, raising concerns about capability degradation.
OpenAI cut inference costs by 50% for non-logged-in users using undisclosed optimization (quantization, cache, or routing), potentially reducing model quality for low-value tasks.
Base44 launches a narrowly fine-tuned model for web apps, slashing inference spend by 75% while maintaining performance—a playbook similar to Cursor's Composer.
Claude Sonnet 5 outperforms Opus 4.8 on economically valuable work (GDP Val) but is token-inefficient and costly, best used as a sub-agent for Fable 5 rather than standalone.
Host recommends using Fable 5 for strategic thinking and writing—its steerability and resistance to sycophancy outshine GPT-5.5 and Opus 4.8, especially on instruction-following tasks.
"The first rule of Fable Club is you do not ask too many questions about what exactly Anthropic agreed to that they weren't doing before, and you enjoy your access."
— Miles Brundage
"There has been a significant breakthrough in architecture, specifically around memory efficiency, not by one of the big labs, but by a team that was spun out of OpenAI. They will probably announce it soon."
— Andrew Curran
"If you use this thing the same way you've been using old models, you're gonna have a bad time."
— Ben Davis (on Sonnet 5)