PulseAugur
EN
LIVE 09:33:15

LLMs fail geometric stability tests in chess evaluation, paper finds

By PulseAugur Editorial · [1 sources] ·

A new research paper published on arXiv analyzes the geometric stability of six large language models (LLMs) when evaluating chess positions. The study found that all tested models, including Claude Sonnet 4.5, DeepSeek Chat, Gemini 2.5 Flash, GPT-5.1, Grok 4.1 Fast, and Kimi K2 Turbo, struggled with evaluating rotated pawnless positions, often misclassifying legal positions as illegal. While GPT-5.1 showed the lowest mean stability error and Claude Sonnet 4.5 the lowest accuracy error, the paper argues that standard accuracy metrics can mask significant stability failures, particularly in transformations like rotation. AI

IMPACT Highlights potential weaknesses in LLM reasoning and consistency beyond simple accuracy metrics.

RANK_REASON Research paper published on arXiv analyzing LLM performance on specific task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

LLMs fail geometric stability tests in chess evaluation, paper finds

How we ranked this

Signal score
16 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
Research paper published on arXiv analyzing LLM performance on specific task. [lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
paper, safety
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

Full methodology in our editorial standards.

COVERAGE [1]

  1. arXiv cs.AI TIER_1 English(EN) · Qingya Hu, Ruifeng Cao, Weiqi Wang, Xidan Song ·

    Beyond Accuracy: A Geometric Stability Analysis of Large Language Models in Chess Evaluation

    arXiv:2512.15033v2 Announce Type: replace Abstract: Large language models (LLMs) are usually evaluated in chess by how closely their position evaluations agree with a strong engine, but such agreement does not show that a model evaluates equivalent positions consistently. We appl…

RELATED ENTITIES

RELATED TOPICS