Model Lab Daily
Daily benchmarks, eval results, and lab notes from the people training the next generation of models.
Latest Analysis
-
ICLR 2026 Best Paper Awards Highlight Diverse Advances in AI Research
ICLR 2026’s Best Paper Awards honored diverse innovations: Spherical DYffusion’s climate model, Mila’s world models, and a notable systems paper…
-
ICLR 2026 Honors Pioneering 2016 Layer Normalization Paper with Award
The ICLR 2026 Test of Time Award recognizes the enduring impact of the 2016 Layer Normalization paper by Ba, Kiros,…
-
Spherical DYffusion Projects Century of Climate Patterns in Record Time
UC San Diego and the Allen Institute for AI unveil Spherical DYffusion, a groundbreaking climate model that forecasts 100 years…
-
Neuromorphic Chips Surpass Supercomputers in Physics Simulation Efficiency
Neuromorphic chips have achieved a remarkable milestone, solving complex physics simulations with significantly less energy compared to traditional supercomputers.
-
OpenAI’s GPT-5.4 Thinking Model Achieves Human Expert Parity on GDPVal Benchmark
OpenAI’s GPT-5.4 ‘Thinking’ model has scored 83.0% on the GDPVal benchmark, marking the first time a public AI system has…
-
EU Commission Proposes Mandatory ‘Ethical Readiness’ Audits for High-Risk AI Systems
The European Commission unveiled a draft directive this morning requiring independent third-party audits of high-risk AI systems before deployment, with…