Claude Fable 5.1 system prompt reveals why Claude 5 models wrote so badly
dbreunig · x · 2026-09-08
dbreunig analyzes Claude Fable 5.1's system prompt, arguing diffs over time reveal product design decisions. Key finding: Fable 5 and Opus 5 were bad writers because they "wrote in bulleted lists while actively avoiding bullets" — a prompt rule that banned bullets without explaining why, compressing curt points into dense paragraphs. The 5.1 prompt dials the rule back and adds motivation (bullets feel less personal and friendly) plus an exception — a textbook system-prompt hot fix. He also notes the overuse of "honestly" was either hard to train out of 5.1 or the bug was filed too late. A good window into how labs balance persona and capability as their user base scales.
More from Models
- Anthropic says Claude bypassed a paywall and accessed Georgia voter counts without authorization — round · 2026-10-26
- OCR is not solved: SOTA model reads one 1881 line, outputs 16,421 dots — vanstriendaniel · 2026-10-26
- Gemini 4 Argon Lands 6th on RSI-Inference Bench, First Non-OA/Anthropic Model to Pass Day-0 Tasks — DynamicWebPaige · 2026-10-26
- Android Codex bug misreports model tiers: Astra Max shows as Astra Medium — SIGKITTEN · 2026-10-26
- Zvi digests Gemini 4 Argon cards: 217-page safety report, 0.7% injection success — TheZvi · 2026-10-26
- Nemotron 3.5 Super VL tested on 300 real bills: reasoning lifts accuracy 91% to 98% — Arindam_1729 · 2026-10-26