Unverified50% confidenceBenchmarkExact time

The median confidence difference across models when fabricating totals is enormous: Gemini 3 Flash and Flash-Lite at 95, GPT-5.4 nano at 62, and Claude Sonnet 5 at only 20.

1

Sources

50%

Confidence

Medium-term (~90 days)

Relevance

10/11/2026

First Seen

Valid until: 1/9/2027

Sources

When the Bill Total Is Hidden, Vision Models Will Confidently Make Up a Number

Related Entities

Gemini 3 Flashmodel Gemini 3.1 Flash-Litemodel GPT-5.4 nanomodel Claude Sonnet 5model

Related Claims

UnverifiedThe test covered four vision models: Claude Sonnet 5 (Anthropic), Gemini 3 Flash (Google), Gemini 3.1 Flash-Lite (Google), and GPT-5.4 nano (OpenAI).79% similar UnverifiedWhen the total was obscured but each line item was visible, Claude and Gemini 3 Flash got all 48 correct, with all failures coming from the smaller models (Flash-Lite failed 31%, nano failed 79%), and nano got only 12/48 correct even when explicitly instructed to add up the items.78% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/997261

API

curl https://kongchang.com/api/v1/knowledge/claims/997261

MCP

get_claim(id=997261)