Consumer AI
Meta’s Muse Wonder Full Ties Claude Sonnet 5.5 on AI Benchmarks
Meta Superintelligence Labs has launched two proprietary frontier models that match or beat most of OpenAI's and Anthropic's…
AI Agents
Sai Agent Hits 73% on OSWorld 2.0 in Key Automation Benchmark
Sai, a computer-use agent built by Simular, hit 73% on OSWorld 2.0 this week, beating GPT-5.6 Sol (62.57%,…
Enterprise AI
5 AI Benchmarks Fail Real-World Enterprise Tests
Enterprise AI teams are spending heavily on models that ace standardised benchmarks, then watching them fail in production.…
AI Research
AI Research Agents Generate Polished Pseudoscience, New Benchmark Reveals
When prompted to support a perpetual motion machine or endorse a mystical energy theory, seven leading AI auto-research…
AI Research
Mistral Orion Rocks AI
Mistral AI has quietly assembled something its better-funded rivals haven't: a full-stack enterprise AI platform built specifically for…
AI Research
Stanford AI Index Reveals Benchmark Saturation and Score Manipulation Tactics
Every major AI lab claims state-of-the-art performance. The 2024 Stanford AI Index Report has now formalised what researchers…
AI Agents
ReplicatorBench Shows LLM Agents Fail to Replicate Half of Social Science Studies
A new benchmark called ReplicatorBench puts a hard number on something researchers have long debated: AI agents can…
AI Research
NeurIPS 2026 Unveils “Evaluations” Track
One of AI's most prestigious conferences just admitted the field has a reliability problem. NeurIPS — the Neural…