Archive
Articles
-
Liquid AI releases open-weight d1 models for edge decisions
Liquid AI’s d1-3B and experimental d1-omni-600M return structured decisions rather than conversational answers. Downloadable weights and inference support are available, but performance claims, multimodal reliability and commercial licensing require qualification.
-
Open Medical-LLM Leaderboard: What Medical AI Scores Leave Unanswered
The medical AI benchmark’s public code and historical results remain accessible, but current rankings are unverified. Mixed protocols and knowledge-focused tests also limit what its scores can establish.
-
AnyLanguageModel: A shared Swift API for local and remote LLMs
AnyLanguageModel brings a Foundation Models-style Swift interface to multiple local and remote backends. Its 2025 introduction now needs to be read alongside breaking changes and Apple’s expanded native framework.
-
NVIDIA reports two gold-level results from specialized Nemotron 3 systems
NVIDIA reports gold-level scores from two specialized Nemotron 3 systems. The IOI run was unofficial, and both results relied on more than fine-tuning alone.
-
Is it agentic enough? Hugging Face’s harness tests models on your tooling
Hugging Face’s agent-eval lets teams test agents against their own libraries. Its author-reported results show why faster runs, fewer tokens and matching answers must be assessed separately.