Benchmarks are shifting toward action, interaction and professional use while exposing gaps between claimed and real capabilities. End-to-end tasks such as business intelligence still show frontier LLM accuracy below 50%.
A study on thousands of tutoring-session transcripts shows that using disagreement among multiple LLMs to target expert attention shortens codebook revision from months to days. Among three methods, Rationale Labeling reaches 64.9% accuracy against 5...
PsyAgentBench reruns five classic psychological experiments on open-weight LLMs under named, blind, canonical, and counterfactual conditions. The results show that effects like Asch conformity depend on the prompt and safety filter, not on a stable b...
The Hugging Face page for XiaomiMiMo/MiMo-V2.6-Flash-RL appeared with minimal details. For teams planning on-prem deployments, this is a reminder: without VRAM, quantization, context window, and throughput specs, a name is not enough. Independent ver...
Prism-LM released Bonsai 2 models based on Qwen3.8, evaluated with the same methodology used for the rest of the comparison. The result, roughly 91.5% on a composite benchmark, is a relative reference point, not an absolute figure. Using Prism's fork...
Bonsai's own documents show Ternary Bonsai 2 27B scoring 52.8 on Terminal-Bench 2.1 and 60.8 on SWE-bench Verified, roughly three quarters of the full-precision model used as comparison. Yet the competitor name appears wrong, and the 98.2% retained-c...
A systematic mapping of 14,767 arXiv preprints published between January 2022 and August 2026 shows how LLM benchmarks are shifting toward action, interaction, and professional applications. Model participation is asymmetric: LLM-based scoring grows,...
