AI News
Latest news and trends from the world of artificial intelligence
Deep Cogito goes big, releasing 4 new open source hybrid reasoning models with self-improving ‘intuition’
Deep Cogito, a San Francisco-based AI research startup, has released four new open-source hybrid reasoning models that aim to improve reasoning effectiveness over time. The models range from 70 billion to 671 billion parameters and are available under a mix of open licensing terms. They include dense and MoE models, designed for different needs. The models are designed as hybrid reasoning systems, able to reflect internally before answering, with reflection baked into the training process. Over time, they learn which lines of thinking matter. The 671B MoE model matches or exceeds leading open models on benchmarks. The models are available on Hugging Face and for local usage, or through APIs from Together AI, Baseten and RunPod. Deep Cogito claims to have trained all eight of its Cogito models for under $3.5 million. The company's roadmap involves running models, learning from their reasoning traces, and repeating the loop, with each model becoming a stepping stone for the next. Deep Cogito's models are open source.
Hard-won vibe coding insights: Mailchimp’s 40% speed gain came with governance price
Mailchimp achieved a 40% speed gain with AI coding tools but learned that governance and human expertise are still essential.
Apple plans to ‘significantly’ grow AI investments, Cook says
Apple is open to M&A to accelerate its AI strategy, and says it's made seven acquisitions this year so far.
Amazon DocumentDB Serverless database looks to accelerate agentic AI, cut costs
Amazon DocumentDB Serverless database brings automatic scaling to MongoDB-compatible document databases, ideal for unpredictable demand scenarios, especially with the rise of AI agents. AWS claims it can reduce costs by up to 90% compared to traditional provisioned databases for variable workloads.
Cohere's new vision model can process images, diagrams, PDFs, and other types of visual data
Cohere's new Command A Vision model is designed to handle images, diagrams, PDFs, and other types of visual data. Cohere says the model outperforms GPT-4.1, Llama 4 Maverick, Pixtral Large, and Mistral Medium 3 on standard vision benchmarks.
Google adds image-to-video and Veo 3 Fast to the Gemini API
Google has rolled out the Veo 3 Fast version and image-to-video support through its API. Veo 3 Fast is engineered for speed and cost efficiency, aimed at developers who need to iterate quickly or generate content at scale. Both Veo 3 and Veo 3 Fast accept text and image prompts, generate videos at 720p and 24 fps, and output eight-second clips by default. The new image-to-video feature allows users to combine a single image with a text prompt to generate dynamic videos with audio.
You’ve heard of AI ‘Deep Research’ tools…now Manus is launching ‘Wide Research’ that spins up 100+ agents to scour the web for you
Manus is launching ‘Wide Research’ that spins up 100+ agents to scour the web for you, taking a different approach than 'Deep Research' tools offered by other AI providers.
BFL and Krea release FLUX.1 Krea: Open image model designed for realism
Black Forest Labs and Krea AI have released FLUX.1 Krea [dev], an open text-to-image model designed to generate more realistic images with fewer of the exaggerated, AI-typical textures.
Google launches Opal: Build AI mini-apps with natural language, no coding needed
Google is rolling out Opal, a new experimental tool that lets users build AI-powered mini-apps with simple natural language prompts, no coding required.
China raises security concerns over Nvidia's H20 chips
China is warning Nvidia about potential security risks related to possible tracking features in its H20 chips.
Archive
About Sources
AI news are automatically downloaded from various sources and translated using AI. Updates occur twice a day.