Google has introduced a new ultra-light model called Gemma 3 270M ideal for smartphones and other on-device use cases. The open-source model is power-efficient and small enough to run locally in the absence of an Internet connection, as Google demonstrated in internal tests using a Pixel 9 Pro SoC. With just 270 million parameters, Gemma 3 270M is a fraction of the size of flagship LLMs, which typically have billions of parameters. While Google’s new model was not made for complex conversational use, it is “designed from the ground up for task-specific fine-tuning with strong instruction-following.” Continue reading Google Says New Gemma 3 Is Ideal for Mobile, Edge Devices

Google DeepMind has unveiled Genie 3, a world-building model that uses text and image prompts to generate 3D environments in real time. Still in research preview, Genie 3 can output “several minutes” of video that can be navigated in real time at 24fps and a resolution of 720p. Because it remembers the rules of the world it creates, Genie 3 allows agents to predict how the environment evolves and how actions affect it. Google says world models are “a key steppingstone” to artificial general intelligence, or AGI, since they can train AI agents in “an unlimited curriculum of rich simulation.” Continue reading Genie 3 World Model Produces Minutes of Video in Real Time

AI audio firm ElevenLabs has launched Eleven Music, which lets businesses and individuals generate studio-caliber music using natural language prompts. Users can generate tracks in any genre or style, even adding vocals in different languages. The Eleven Music model was developed in partnership with music licensing firm Merlin and independent publisher Kobalt. Artists and songwriters from the two groups will participate in the development of Eleven Music Pro, a subsequent model planned for release in the coming months. The company says it built-in guardrails to protect rightsholders. Continue reading ElevenLabs Debuts Eleven Music with Kobalt, Merlin Backing

Alibaba’s Qwen team has launched Qwen3-Coder, which it calls its “most agentic code model to date.” While it will be made available in multiple sizes, the most powerful variant — Qwen3-Coder-480B-A35B-Instruct — is being released first. The 480 billion parameter mixture-of-experts model has 35 billion active parameters supporting a context length of 256,000 tokens natively and 1 million tokens with extrapolation methods for “exceptional performance in both coding and agentic tasks,” explains the group, which claims the quasi-open source model has agentic coding, agentic browser use, and agentic tool use comparable to Anthropic’s proprietary Claude Sonnet 4. Continue reading Alibaba Is Rolling Out Its ‘Most Agentic Code Model to Date’

Oracle and OpenAI have confirmed the $30 billion per year AI data center contract reported earlier this month. The deal will provide OpenAI’s Stargate project with 4.5 gigawatts of additional data center capacity in the U.S. Oracle is a Stargate partner and has been working in partnership with OpenAI on the Stargate I site, coming online in Abilene, Texas. “This additional partnership with Oracle will bring us to over 5 gigawatts of Stargate AI data center capacity under development, which will run over 2 million chips,” OpenAI explains. The “investment will create new jobs, accelerate America’s reindustrialization, and help advance U.S. AI leadership.” Continue reading OpenAI and Oracle Confirm $30B Annual Data Center Contract

AI is now part of every paid Slack subscription, and the platform continues building out its agentic OS with new features including enterprise search connectors, writing assistance, and contextual definitions embedded in the app. The Enterprise+ plan, which scales AI company-wide, now includes AI for drafting documents and answering questions using information stored in Slack chats and connected apps. Business+ is adding recaps, translations, workflow generation and AI-powered search, while the entry-level Pro tier offers AI summarization for conversations in channels, threads and huddles, making it easier to stay current on communications. Continue reading Slack to Roll Out New AI Features for Enterprise Collaboration

The European Union has published a General Purpose AI (GPAI) Code of Practice designed to help companies comply with the AI Act, which includes copyright protections and transparency requirements for advanced models. The Code of Practice bans training models on unauthorized materials and says companies must comply with copyright-holder requests to omit work from datasets. Developers are required to provide documentation describing the features of their AI models. The AI Act began taking effect in August 2024 and is being implemented gradually, with key transparency, governance and privacy provisions coming into force next month. Continue reading EU Releases AI Practices Code to Help with Legal Compliance

AI startup Runway has a new tool called Game Worlds that lets users generate simple video game worlds using images and text-based prompts. At the moment, Runway Game Worlds can only help generate simple text-based interactive adventures that include pictures, but the company has plans to enable more complex game creation by the end of the year. Runway CEO Cristóbal Valenzuela says the company is interested in partnering with video game companies who are willing to provide game data that can be used to train the company’s models in exchange for generative capabilities. Continue reading Runway AI Intros Game Worlds Generator in Limited Preview

Creative Commons, the non-profit that pioneered sharing content through permissive licensing, is launching CC Signals, a framework to signal permissions for content use by machines in the age of artificial intelligence. “They are both a technical and legal tool and a social proposition: a call for a new pact between those who share data and those who use it to train AI models,” says Creative Commons CEO Anna Tumadóttir, noting the signals are “based on a set of limited but meaningful options shaped in the public interest.” The framework is designed to bridge the openness of the Internet with AI’s insatiable demand for training data, according to Creative Commons. Continue reading Creative Commons Introduces New Licensing Platform for AI

OpenAI is working to build a social network that will compete against Elon Musk’s X and Meta’s Instagram, reports say. Though still in the early stages, the project is revolving around an internal prototype that is said to involve a social feed that leverages ChatGPT’s image generator. It’s unclear if an OpenAI social app would be standalone or integrated with ChatGPT, but either way it would most likely heighten the competition between rivals Musk and OpenAI CEO Sam Altman, who recently fended off an unsolicited offer by Musk to purchase his company for $97.4 billion. Continue reading OpenAI Reportedly Has Prototype for Its Own Social Network

A new open-source code reasoning model called DeepCoder-14B-Preview has hit the market. Built atop DeepSeek-R1 and Qwen2.5 using reinforcement learning (RL), it aims to provide more flexibility by combining high-performance code generation with reasoning capabilities for real-world applications. Its performance is said to be comparable to OpenAI’s o3-mini, “but with a smaller footprint,” say its developers, the research-driven AI companies Together AI and Agentica. “We democratize the recipe for training a small model into a strong competitive coder,” explains Together AI. Continue reading Researchers Debut Preview of DeepCoder Reasoning Model

San Francisco-based AI startup Deep Cogito has released five AI models in preview, making them available under an open-source license agreement. The models come in sizes 3B, 8B, 14B, 32B and 70B, with plans to release 109B, 400B and 671B versions in the weeks and months ahead. As for the current models, “each outperforms the best available open models of the same size, including counterparts from Meta, DeepSeek and Alibaba, across most standard benchmarks,” Deep Cogito claims, noting that the 70B model in particular “outperforms the newly released Llama 4 109B MoE model.” Continue reading Deep Cogito Is Out of Stealth with Hybrid Reasoning Models

OpenAI has activated the multimodal image generation capabilities of GPT-4o, making it available to ChatGPT users on the Plus, Pro, Team and Free tiers. It replaces DALL-E 3 as the default image generator for the popular chatbot. GPT-4o’s accuracy with text, understanding of symbols and precision with prompts combined with well multimodal capabilities that allow the model to take cues from visual material have transformed its image capabilities from largely unpredictable to “consistent and context-aware,” resulting in “a practical tool with precision and power,” claims OpenAI. Continue reading OpenAI Delivers Native GPT-4o Image Generator to ChatGPT

Elon Musk’s xAI has acquired generative video startup Hotshot to bring motion imaging to Grok 3. Released in February, Grok 3 adds Deep Search and Thinking and improved on its predecessor’s still imaging capabilities, but lacks generative video, a much-requested feature — one that could make Grok a freestanding competitor to OpenAI’s individual offerings: ChatGPT for text, Sora for video, and DALL-E for images. “Cool AI video coming soon!” was Musk’s comment to Hotshot’s acquisition announcement on the networking platform. Hotshot can generate clips of up to 10-seconds at 1280×720 pixels. Continue reading With Hotshot Purchase, xAI to Bring Generative Video to Grok

OpenAI is urging the Trump Administration to declare AI training fair use, seeking unfettered access to copyrighted material for the purpose of educating models. The company is also asking for relief from state AI rules and more permissive AI export rules in a response to President Trump’s call for a U.S. “AI Action Plan.” The deadline to submit responses to the National Science Foundation and Office of Science & Technology Policy (OSTP) request for information (RFI) regarding the plan was Saturday. Google also publicized its response, which largely echoed OpenAI’s points. Continue reading OpenAI and Google Press for Relief on Copyright, State Laws