Moonshot AI · Benchmark
PerceptionBench
3,000 verified questions isolating 10 atomic perceptual skills -- counting, depth/3D, OCR, fine-grained recognition, hallucination detection and more -- built by decomposing real model failures across 42 benchmarks (60 percent) plus newly authored items (40 percent). Across 16 frontier MLLMs tested, none clears 60 percent accuracy.
Open the source ↗Similar Research & Safety releases
- Real-SWE
Specific Labs · Enterprise Coding Benchmark · Week 37, 2026
- SimReady Assets
NVIDIA · Simulation asset dataset · Week 45, 2026
- Dream-RSI
Google, DeepMind, Maryland & Virginia · Recursive Self-Improvement · Week 39, 2026
- AlphaGenome Atlas
Google DeepMind · AlphaGenome Atlas · Week 37, 2026
- Male Fruit Fly Connectome
Google Research and HHMI Janelia · Fly Brain Connectome · Week 36, 2026
- WeatherNext 3
Google DeepMind · WeatherNext 3 · Week 36, 2026
More from Moonshot AI
- Kimi K3.1 Pro
Moonshot AI · Open-Weight Frontier · Week 44, 2026
- Kimi K3 (weights)
Moonshot AI · MoE Architecture · Week 31, 2026
- Kimi K3
Moonshot AI · MoE Architecture · Week 29, 2026
- Kimi Code
Moonshot AI · Coding Agents · Week 24, 2026
- Kimi K2.6
Moonshot AI · Code Generation · Week 17, 2026
- Kimi-K2.5
Moonshot AI · Reasoning · Week 5, 2026
Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.