Hugging Face Launches POCKET: 35-Billion-Parameter Model for iPhone and PC
Hugging Face introduces POCKET, a 35-billion-parameter model that runs on iPhones and PCs without a GPU, achieving 2.7× faster CPU performance than a leading 1-bit 27B model.
Model Release
459 articles
New AI models as they ship — large language models, multimodal systems, and open-weight releases. Capabilities, context windows, pricing, and availability, summarized so you can compare what is actually new.
LATEST NEWS
-
-
U.S. Treasury Warns of Sanctions Over Moonshot's Fable Distillation
U.S. Treasury Secretary Scott Bessent warned of sanctions against Moonshot after the White House accused it of distilling Anthropic’s Fable model, a process that could infringe on intellectual property rights.
-
Cisco Releases Small Open Cybersecurity Models for Cost-Effective Vulnerability Detection
Cisco's Antares-350M and Antares-1B models detect 150 times more vulnerabilities per dollar than large AI agents like Cognition's Devin Security Swarm.
-
Google AI Introduces Gemini Intelligence on Samsung Devices
Google AI announced Gemini Intelligence 1 will debut on Samsung Galaxy Z Fold8 Ultra, Fold8, and Flip8 devices in 2026, expanding task automation to over 40 apps.
-
Meta Testing AI Bedtime Story App Called StoryKit
Meta is piloting an AI storytelling app called StoryKit that generates children's stories with custom characters, settings, and lessons, according to a TechCrunch report.
-
NVIDIA Releases Cosmos 3 Edge Model on Hugging Face
NVIDIA announced the release of Cosmos 3 Edge, a 4-billion-parameter model designed for edge computing, on July 20, 2026.
-
Google Launches Three New Gemini Flash Models Amid Pro Delay
Google released three new Gemini Flash models, including a cybersecurity-focused variant, but its flagship 3.5 Pro remains delayed, raising concerns about its competitive edge.
-
Google releases faster and cheaper Gemini 3.6 Flash
Google announced Gemini 3.6 Flash, a faster and cheaper version of its AI model, with 17 percent fewer token usage than its predecessor.
-
Google Releases Three New Gemini Models But No 3.5 Pro
Google DeepMind launched three new Gemini models on Tuesday, including a cost-effective variant with up to 17% lower token usage than its predecessor.
-
Alibaba's Qwen-Image-3.0 Renders Full Infographics and Ten-Pixel Text
Alibaba's Qwen-Image-3.0 can generate legible ten-pixel text and full infographic grids in a single pass, according to the company.
-
DeepMind Launches Gemini 3.5 Flash Cyber for Cybersecurity
DeepMind announced Gemini 3.5 Flash Cyber, a lightweight cybersecurity model, on July 21, 2026, to help find and fix vulnerabilities more efficiently.
-
Z.ai's GLM 5.2 Model Closes Coding Gap With U.S. Rivals
Z.ai's GLM 5.2 model outperformed U.S. counterparts in coding benchmarks, narrowing the performance gap in key areas.