Unverified50% confidenceBenchmarkExact time

Gemma 4 31B, Gemma 4 26B-A4B, Gemini 3.7 Flash, GPT-6-sol, and GLM-5 achieved 100% fidelity in this benchmark.

1

Sources

50%

Confidence

Medium-term (~90 days)

Relevance

10/11/2026

First Seen

Valid until: 1/9/2027

Sources

I Planted 962 Typos to Test AI Verbatim Quoting — Not One Was 'Corrected.' The Real Failures Were All About Sentence Boundaries

Related Entities

Gemma 4 31Bmodel Gemma 4 26B-A4Bmodel Gemini 3.7 Flashmodel GLM-5model

Related Claims

UnverifiedA post on Reddit claimed that the model GPT-6.1 Sol (max) achieved a score of 100% on the FrontierMath Tier 4 benchmark.70% similar UnverifiedGPT-6 Astra achieved 100% accuracy on the eight-needle benchmark in the 256K–512K token range.69% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/1000255

API

curl https://kongchang.com/api/v1/knowledge/claims/1000255

MCP

get_claim(id=1000255)