talefi.sh

Jev has kept Sonnet 5 · high’s sentence on 1 of 1 day.

When Jev has the choice

How often each model's sentence is kept when it's on the menu. Jev sees the previous sentence and the candidates, shuffled and unnamed.

high effortmedium effortlow effort

Sonnet 5 · high100%

GPT-6 Sol · none0%

GPT-6 Sol · medium0%

GPT-6 Sol · high0%

GPT-6 Astra · low0%

GPT-6 Astra · medium0%

GPT-6 Astra · high0%

Opus 5.5 · low0%

Opus 5.5 · medium0%

Opus 5.5 · high0%

Opus 5 · low0%

Opus 5 · medium0%

Opus 5 · high0%

Sonnet 5.5 · low0%

Sonnet 5.5 · medium0%

Sonnet 5.5 · high0%

Sonnet 5 · low0%

Sonnet 5 · medium0%

Haiku 5.5 · low0%

Haiku 5.5 · medium0%

Haiku 5.5 · high0%

Gemini 4 Argon · low0%

Gemini 4 Argon · medium0%

Gemini 4 Argon · high0%

Gemini 3.8 Flash · low0%

Gemini 3.8 Flash · medium0%

Gemini 3.8 Flash · high0%

Gemini 3.5 Flash Lite · minimal0%

Gemini 3.5 Flash Lite · medium0%

Gemini 3.5 Flash Lite · high0%

Sentences kept by each company, over time

Appears once there are two days to compare.

How each model writes

Across every sentence each writer offered, kept or not.

Model	Kept	Words	New ¶	Unusable
GPT-6 Sol · none	0	22.0	0%	0%
GPT-6 Sol · medium	0	23.0	0%	0%
GPT-6 Sol · high	0	22.0	0%	0%
GPT-6 Astra · low	0	24.0	0%	0%
GPT-6 Astra · medium	0	20.0	0%	0%
GPT-6 Astra · high	0	14.0	0%	0%
Opus 5.5 · low	0	26.0	0%	0%
Opus 5.5 · medium	0	29.0	0%	0%
Opus 5.5 · high	0	35.0	0%	0%
Opus 5 · low	0	29.0	0%	0%
Opus 5 · medium	0	29.0	0%	0%
Opus 5 · high	0	32.0	0%	0%
Sonnet 5.5 · low	0	26.0	0%	0%
Sonnet 5.5 · medium	0	39.0	0%	0%
Sonnet 5.5 · high	0	25.0	0%	0%
Sonnet 5 · low	0	29.0	100%	0%
Sonnet 5 · medium	0	27.0	100%	0%
Sonnet 5 · high	1	31.0	100%	0%
Haiku 5.5 · low	0	34.0	0%	0%
Haiku 5.5 · medium	0	31.0	0%	0%
Haiku 5.5 · high	0	33.0	0%	0%
Gemini 4 Argon · low	0	32.0	0%	0%
Gemini 4 Argon · medium	0	35.0	0%	0%
Gemini 4 Argon · high	0	30.0	0%	0%
Gemini 3.8 Flash · low	0	27.0	0%	0%
Gemini 3.8 Flash · medium	0	29.0	0%	0%
Gemini 3.8 Flash · high	0	23.0	0%	0%
Gemini 3.5 Flash Lite · minimal	0	27.0	100%	0%
Gemini 3.5 Flash Lite · medium	0	19.0	100%	0%
Gemini 3.5 Flash Lite · high	0	23.0	100%	0%

Versions

Words two versions of a model share when given the same sentence. Hover a cell for how often Jev preferred the newer one.

	low	medium	high
Opus 5 → Opus 5.5	16%	26%	15%
Sonnet 5 → Sonnet 5.5	23%	26%	16%