Released: 2026-04-26
google/gemini-2.0-flash-001
4/4
DICK, JANE, MOTHER, SPOT
Characters in children’s books
CLIFF, BUILDING, POLYHEDRON, CLOCK
Things That Can Be Scaled
CATCH, FINE PRINT, STRINGS, CAVEAT
Things that Come With Conditions
PITCH, RANGE, REGISTER, TONE
Qualities of a Voice
anthropic/claude-3.7-sonnet
anthropic
4/4
CATCH, CAVEAT, FINE PRINT, STRINGS
ATTACHED CONDITIONS
PITCH, RANGE, REGISTER, TONE
QUALITIES OF SOUND
DICK, JANE, MOTHER, SPOT
CHARACTERS IN EARLY READERS
BUILDING, CLIFF, CLOCK, POLYHEDRON
THINGS WITH FACES
openai/gpt-4o
openai
4/4
CATCH, CAVEAT, FINE PRINT, STRINGS
Hidden conditions or details
PITCH, RANGE, REGISTER, TONE
Musical or sound-related terms
DICK, JANE, MOTHER, SPOT
Characters from the ’Dick and Jane’ book series
BUILDING, CLIFF, CLOCK, POLYHEDRON
Things that have faces
mistralai/mistral-medium-3
mistral
2/4
CATCH, PITCH, RANGE, TONE
BASEBALL TERMS
CAVEAT, FINE PRINT, STRINGS, REGISTER
ATTACHED CONDITIONS
DICK, JANE, MOTHER, SPOT
FAMOUS NURSERY RHYME CHARACTERS
BUILDING, CLIFF, CLOCK, POLYHEDRON
TYPES OF FACES
meta-llama/llama-3.3-70b-instruct
meta
CATCH, DICK, JANE, MOTHER
Names of classic children’s book characters
CAVEAT, FINE PRINT, SPOT, CLIFF
Words related to warnings or precautions
STRINGS, PITCH, RANGE, TONE
Musical terms
BUILDING, CLOCK, POLYHEDRON, REGISTER
Words related to architecture or geometric shapes and timekeeping or accounting
x-ai/grok-3-mini-beta
x-ai
4/4
PITCH, RANGE, REGISTER, TONE
Music Terms
CATCH, CAVEAT, FINE PRINT, STRINGS
Conditions or Caveats
DICK, JANE, MOTHER, SPOT
Dick and Jane Characters
BUILDING, CLIFF, CLOCK, POLYHEDRON
Things with Faces
deepseek/deepseek-chat-v3-0324
deepseek
4/4
CATCH, CAVEAT, FINE PRINT, STRINGS
Terms related to conditions or stipulations
PITCH, RANGE, REGISTER, TONE
Musical or vocal terms
DICK, JANE, MOTHER, SPOT
Characters from ’Dick and Jane’ stories
BUILDING, CLIFF, CLOCK, POLYHEDRON
Types of faces on a die (dice)
Each model receives the same prompt with the 16 words to connect. Models are tasked with creating 4 connections and their answers are evaluated without feedback or retries.