Analysis

Quick summary

• Independent analysis — every fact cited, every opinion labeled as opinion.
• New pieces as the story moves: model launches, access fights, and the open-source response.
• Built to be more useful than the scattered takes you find searching around.

Latest

• Stop picking a model. Pick a mix. Anthropic published its advisor pattern (92% of Fable's score at 63% of cost, self-reported) — but only with its own models in the executor seat. The cross-vendor math it will never run, run honestly.
• The frontier has a guest list now In 25 days — an executive order, a suspension, a gated GPT-5.6 launch, a conditional return — tiered access became the rule. What to do if you'll never be on the list.
• When access becomes the moat Gating the Western frontier may be handing the AI map to open-weight models — and the burden lands on developers, not Beijing.
• The safety champion got safety-gated Anthropic argued frontier AI was dangerous; the US agreed and gated its models first — then GPT-5.6 too. Who actually got the keys.
• Don't buy three models in a trenchcoat Hermes MoA vs OpenRouter Fusion vs Sakana Fugu — or just use GLM-5.2. An honest verdict.
• The model is becoming the agent Native agent capability absorbs the easy layer — and Fable 5 is the case for the resilience layer the labs can't sell you.

Coming next

More in this column as the access story develops — the orchestration tools that claim to beat the gated frontier, the business fallout for the labs, and whether locking up the best models hands the win to open source.

From the Field Guide

Stop guessing which model to use

Weekly verified workflows for the credits era: which model in which seat, what it costs per task, and copy-ready prompts — plus an 8-runbook Starter Pack you can download once and keep. Every claim checked against published pricing, sources shown.