Claude AI interface showing longer, more detailed answers after "Honestly" and "Frankly" cuts.

Editorial illustration for Claude's Fable 5.1 Cuts "Honestly" and "Frankly" as Answers Grow 30% Longer

Claude's Fable 5.1 Cuts "Honestly" and "Frankly" as...

By AI Daily Post Edited by Brian Petersen, Editor-in-Chief

September 10, 2026 • 3 min read

Anthropic's Claude Fable 5.1 talks differently than its predecessor did, and Arena.ai has the numbers to show it. The research group ran tens of thousands of high-reasoning Text Arena outputs through a comparison of Fable 5 and Fable 5.1, tracking word choice, sentence habits, and length. What they found is a model that hedges less, agrees less reflexively, and pads its answers more.

Median response length jumped 30 percent, from 319 words to 414. That's still shorter than Opus 5, which averages 525 words per response, but the gap is closing. At the same time, Fable 5.1 dropped a chunk of the verbal tics that made Fable 5 sound like it was stalling for time or softening a point before making it.

Words like "honestly" and "frankly," along with agreement openers and em dashes, show up far less often. Anthropic hasn't published its own breakdown of the shift, but Arena.ai's dataset gives a clear picture of a model writing longer answers with noticeably less filler holding them up.

Arena.ai analyzed how Anthropic's Claude's writing changed from Fable 5 to Fable 5.1 across tens of thousands of high-reasoning Text Arena outputs. Fable 5.1 uses fewer agreement openers, fewer em dashes, and less wording like "honestly" and "frankly," while its answers have grown longer.

Claude Fable 5.1's language is less "load-bearing" than its predecessor's - THE DECODER

Why this matters

For developers building on Claude, wording shifts like this are a proxy for something harder to measure: how a model's personality gets tuned between point releases. Fewer "honestly," fewer em dashes, fewer agreement openers, that's Anthropic dialing down the conversational filler that made Fable 5 sound like it was hedging or buttering you up before answering. Longer median responses (414 words versus 319) suggest Fable 5.1 is spending more tokens on substance rather than throat-clearing, though it still runs 21 percent shorter than Opus 5's 525-word average.

That gap matters for anyone budgeting on token costs or designing prompts that expect a certain verbosity. Arena.ai's method, mining tens of thousands of high-reasoning outputs for phrase frequency, is a useful reminder that these models have measurable stylistic fingerprints that shift release to release, independent of benchmark scores. If you're fine-tuning prompts around Claude's tone, or comparing it against Opus for a specific use case, don't assume the personality you tested against Fable 5 still holds in 5.1.

Watch for Anthropic's next release to see if this trims further or reverses.