DeepSeek released its V4.1-Flash model, a 552-billion-parameter, multimodal, mixture-of-experts model with a 1-million-token context window

Overview Debates Related claims Topics

The Qwen 3.6-35B A3B language model contains 40 layers, with each layer comprising 256 experts containing 512 neurons each

Selecting the top 64 activating neurons per expert in Qwen 3.6-35B A3B ablates approximately 88% of all neuron activations

In the Qwen 3.6-35B A3B mixture-of-experts model, the router activates eight of the 256 experts at each layer during a forward pass

Individual experts within the Qwen 3.6-35B A3B mixture-of-experts model specialize in one to three specific subjects, such as algebra, geometry, or legal jurisprudence

Canva developed Canva Brush by fine-tuning on top of an existing open-weights vision-language model

Canva announced Canva Brush, a visual language model designed for graphic design creation and editing

Vision-language models can evaluate graphic design quality across metrics including readability, layout, and visual composition when provided structured rubrics

On a Western blot gel electrophoresis task, Claude Opus identified the band threshold at five micrograms instead of one microgram due to visual noise

Vision-language models exhibit poor confidence calibration on biological visual question answering tasks, frequently expressing near 100 percent verbalized confidence on incorrect answers

On a brain tumor histology assay in Handshake AI's Vials benchmark, Claude Opus incorrectly identified panel G as showing advanced necrosis along with panels H and I

When vision-language models generate textual descriptions of laboratory assay images, the descriptions frequently omit critical quantitative details required to solve scientific questions

Google launched Agentic Video Understanding in September 2026

Legora utilizes Google's Gemini models for audio transcription and speaker attribution tasks

Mistral Large 4 uses a sparse mixture-of-experts architecture that activates approximately 50 billion parameters per token

Google revealed the world model Genie 2 in December 2024

Justine Moore of Andreessen Horowitz demonstrated an interactive Lovecraftian simulation and a playable environment titled GTA Waymo generated using Genie 4

Google introduced Genie 4, a multimodal world model that integrates Gemini Omni, Gemini 3.8 Flash, and Google audio models

Google's Genie 3 generated simulated environments autoregressively frame-by-frame with a generation ceiling of 60 seconds

Google uses Genie 4 paired with Gemini models to construct synthetic training environments for robots without physical real-world constraints

Genie 4 generates integrated audio containing procedural sound effects, human speech, and background music within simulated worlds