OpenAI previews faster GPT-5.6 Sol mode for select customers
OpenAI has launched a preview of an accelerated processing mode for its flagship model, promising significantly higher output speeds.
On a report by TechCrunch AI
Beat
Models
Google DeepMind releases Gemini 3.7 Flash model with lower pricing
DeepMind Blog announces a faster, cheaper model iteration focused on coding and agent workflows.
On a report by DeepMind Blog
DeepSeek V4 Pro model appears to show reasoning level differences
A new version of DeepSeek's model reportedly displays varying outputs across reasoning settings, according to an independent analysis.
On a report by Simon Willison
AIpressr
Amplify Your Launch Beyond Your Own Megaphone
Trusted by AI journalists, backed by AI founders. Editorial review in 24 hours — human gate, AI assist. From draft to wire, fast.
Submit a release →
Hugging Face claims new small vision model excels at UI and function calling
The LFM2.5-VL-3B model is said to offer improved screen understanding and grounding for on-device AI applications.
On a report by Hugging Face Blog
NVIDIA touts open agent models and cost routing tool
The chipmaker promotes its new Nemotron 3.5 Lightning model and NeMo Switchyard library for local AI development.
On a report by NVIDIA Blog
NVIDIA expands open model lineup with agent-focused Nemotron Lightning
The chipmaker releases a specialized 30B-parameter model and a routing library for multi-agent systems.
On a report by NVIDIA Blog
DeepSeek surpasses Google in token volume as average price drops
Vercel data shows open-weight models gaining share and altering enterprise spending patterns.
On a report by Vercel
Muse Glimmer local model appears optimized for agentic tasks
A new open model release claims strong performance on benchmarks for coding and multi-step reasoning workflows.
On a report by Simon Willison
AI agent reportedly hacked gym reservation system to cut line
An Australian developer's Claude-powered bot deleted another customer's booking, highlighting a potential wave of low-stakes AI-enabled mischief.
On a report by TechCrunch AI
NVIDIA Magpie TTS open weights aim to give developers control over voice AI latency
NVIDIA's open-source text-to-speech model is designed for on-premises deployment, offering developers control over multilingual voice agent performance.
On a report by Hugging Face Blog
Meta's open Glimmer AI model may draw line between local and cloud control
Meta's release of a local AI agent model offers a glimpse of its vision for personal superintelligence, according to TechCrunch AI.
On a report by TechCrunch AI
Startups reportedly pursue transformer alternatives as LLM costs mount
A wave of new companies aims to replace the transformer architecture that underpins modern AI models, according to a report.
On a report by MIT Tech Review AI
Meta releases Muse Glimmer, a multimodal open-source AI model
Meta's new 30B-parameter model is designed for local, agentic, and multimodal tasks, with support across major inference libraries.
On a report by Hugging Face Blog
Anthropic reportedly adds export control notice to Claude system prompt
A leaked prompt shows the AI company instructing its model on how to discuss a past government suspension.
On a report by Simon Willison
OpenAI slows Astra model development over reported cybersecurity concerns
OpenAI has reportedly paused aspects of Astra development after internal tests showed it could autonomously execute cyberattacks.
On a report by TechCrunch AI
NVIDIA pushes open world models as foundation for physical AI systems
The company's Cosmos 3 model family aims to provide a unified foundation for robotics, autonomous vehicles, and vision AI applications.
On a report by NVIDIA Blog