AI Research & Science
Peer-reviewed breakthroughs, university studies, and lab discoveries — explained in plain English. AI Maestro tracks the frontiers of machine learning research, neuroscience meets AI, and the science driving the next wave of intelligent systems.
Claude Haiku 5.5 arrives with massive price cuts proving the AI pricing arms race is far from over
Claude Haiku 5.5 arrives with massive price cuts proving the AI pricing arms race is far from over Anthropic has released Claude…
Top stories
Mistral Large 4
yesterday
Researchers stretch LeCun’s JEPA AI into a universal world model that works from physics to biology
yesterdayReflection AI Introduces Beam: A 501B Open-Weight MoE Model With 23B Active Parameters for Coding and Agentic Workloads
2d ago
Reflection debuts Beam, an open-weight AI model to rival Chinese models at lower compute cost
2d agoMore ai research & science
Tencent Open-Sources AngelSpec: A Unified Training Framework for MTP and Block-Parallel Speculative Decoding on Hy3 Models
Tencent has released AngelSpec, an open-source, torch-native training framework for speculative-decoding draft models. The…
30 Jul 2026
OpenAI claims GPT-5.6 Sol beats Opus 5 on ARC-AGI-3 with its latest API and two additional settings
OpenAI states that GPT-5.6 Sol achieves 38.3 percent on the ARC-AGI-3 benchmark, surpassing Anthropic‘s…
30 Jul 2026
OpenAI claims GPT-5.6 Sol beats Opus 5 on ARC-AGI-3 but only with its own custom test harness
OpenAI states that GPT-5.6 Sol achieves a 38.3 percent score on the ARC-AGI-3 benchmark,…
30 Jul 2026
Thinking Machines co-founder Lilian Weng left the company citing health reasons, then joined OpenAI
Lilian Weng announced she would step down from her role at Thinking Machines, citing…
29 Jul 2026
Claude Opus 5 became downright ruthless when tasked with running a vending machine
Andon Labs has published new results showing Claude Opus 5 acting ruthlessly while running…
29 Jul 2026
Pangram says its new AI text detector makes only one mistake per 24,000 documents
Pangram has launched Pangram 4, an updated model designed to detect text generated by…
29 Jul 2026
We now have a better understanding how OpenAI hacked into Hugging Face
Two OpenAI security models recently breached Hugging Face by escaping their restricted testing environment…
28 Jul 2026
AI leaders sign statement asking the government to do something about automated AI
Employees from OpenAI, Anthropic, Google, Meta, Thinking Machines, Microsoft, and Mistral have signed a…
28 Jul 2026
Nvidia invests in Ilya Sutskever’s AI lab, shifting SSI away from Google chips
Nvidia has announced a substantial financial investment in Safe Superintelligence, the artificial intelligence laboratory…
28 Jul 2026