Latent Space Frontier AEO Tracker: What Astra Chooses
Latent Space created a Frontier AEO tracker, building on "What Claude Code Actually Chooses," to evaluate model recommendations. The methodology extends AmplifyingAI's approach, running 6 prompt variations over 7 models in 161 categories, from coding agents to AI podcasts. Astra extracts answers, which are then scored using a proprietary AEO system that weights first choices, alternatives, mentions, and negative recommendations. The tracker identifies dominant products, model biases (e.g., Claude Code recommending Fable/Opus), and "soft biases." It also analyzes "consequential flips" in model choices between generations from the same lab, such as Opus to Fable and Sol to Astra, noting differences in source searching and confidence. The analysis includes top cited sources influencing agent recommendations and an examination of top failures.
Original sources
Newsletter Latent Space
AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.