X Article
Field report: no LLM is universally best. Claude Fable 5 is impressive for development but remains expensive, while GPT‑5.6 Sol is my default because its persistence on long tasks offsets the risk of drift.
Public excerpt only. The full article stays on X.
Translated from French.
Read the full article on XArticle in French14 min readArticles on related topics or by the same authors.
X Article excerpt
J’utilise Hermes au quotidien depuis plusieurs mois. Ça m’a donné envie de comprendre l’architecture derrière le produit. Après pas mal de recherches et de tests, j’en ai tiré ce guide. J’espère que ça t’aidera à baisser le coût de tes modèles sur Hermes.
X Article excerpt
Codex prepares the context, Oracle sends a controlled bundle to GPT‑5.6 Sol in ChatGPT Pro, and Codex reviews it before coding and testing. Roles stay separate: Pro reasoning on one side, Codex execution on the other—without the OpenAI API or sending the repository blind.
Translated from French.
X Article excerpt
At Build 2026, Microsoft's ambition goes beyond software: a complete operating system for the agentic era. Its integrated stack spans models, agents, cloud, hardware and energy. Alongside NVIDIA, Microsoft aims to become an industrial AI operator as compute demand explodes.
Translated from French.
