OpenAI set to launch most capable GPT model after delayed rollout ↗
OpenAI is preparing a broad release of GPT-5.6 following additional testing and discussions with US officials. Access had previously been limited to vetted partners amid concerns surrounding cybersecurity and national security.
The flagship Sol model will arrive alongside the cheaper Terra and Luna versions. OpenAI says the family brings improvements across coding, biology, cybersecurity and agentic work - a fairly serious leap, or so it appears.
OpenAI launches GPT-Live voice models that listen and speak simultaneously ↗
OpenAI launched GPT-Live, a family of voice models capable of listening and speaking at the same time. The result should be fewer awkward pauses and more natural interruptions during live conversations.
GPT-Live-1 and the lighter GPT-Live-1 mini are rolling out globally. Simultaneous audio may sound like a minor detail - until voice agents stop behaving like walkie-talkies.
SpaceXAI launches Grok 4.5 model for coding, agentic tasks ↗
SpaceXAI released Grok 4.5, describing it as its smartest model yet for coding and autonomous tasks. It was trained using tens of thousands of Nvidia GB300 chips, with particular attention given to filtering and deduplicating training data.
The model is available through Grok Build, Cursor and SpaceXAI’s developer console. Pricing begins at $2 per million input tokens and $6 per million output tokens - aggressive, conspicuously so.
Mistral launches first robotics model in physical AI push ↗
Mistral unveiled Robostral Navigate, its first model designed for robots operating in factories, warehouses and other physical environments.
The system allows robots to navigate using a single camera, without lidar, complex sensor arrays or multi-camera rigs. It works across different robot suppliers, though it handles navigation rather than object manipulation - baby steps, but with metal feet.
China weighs silicon curtain around sought-after AI models ↗
Chinese authorities have discussed restricting overseas access to advanced domestic AI models, including systems from Alibaba, ByteDance and Z.ai. The move would echo tightening US controls on frontier technology.
That could hurt Western startups and researchers relying on cheaper Chinese open-source models for coding and agentic workloads. In a peculiar twist, an AI border wall could end up strengthening the very US giants it is intended to challenge.
Create videos in seconds with Google Photos’ Video Remix ↗
Google Photos introduced Video Remix, a Gemini Omni-powered tool that reworks ordinary clips through templates and simple prompts.
Users can change backgrounds, apply cinematic lighting or add watercolor, sketchbook and oil-painting effects. The feature is beginning its rollout for eligible Google AI Plus, Pro and Ultra subscribers in selected countries.
Prime Intellect raises $130M Series A to help enterprises build their own AI agents ↗
Prime Intellect raised $130 million at a $1 billion valuation. Radical Ventures led the round, with backing from Nvidia Ventures, Intel Capital, Dell Technologies Capital and several prominent tech founders.
Its platform combines compute access, reinforcement-learning tools and evaluations, allowing companies to train specialised agents without depending entirely on frontier labs. Customers include Ramp and Zapier, while its annualised revenue run rate has reportedly reached $100 million - hardly pocket change.
FAQ
What is GPT-5.6, and when will OpenAI release it?
GPT-5.6 is OpenAI’s next flagship model family, with a wider rollout planned after further testing and discussions with US officials. Access was previously limited to vetted partners because of cybersecurity and national security concerns. The family includes the flagship Sol model, alongside the lower-cost Terra and Luna versions.
What improvements are expected from OpenAI GPT-5.6?
OpenAI says GPT-5.6 will deliver stronger performance across coding, biology, cybersecurity and agentic work. The Sol, Terra and Luna versions are expected to let users choose between maximum capability and lower operating costs. Precise performance will vary according to the model version and the task at hand.
How does GPT-Live make voice AI conversations more natural?
GPT-Live models can listen and speak simultaneously, rather than waiting for one side to finish completely. This should reduce awkward pauses and make interruptions feel more natural during conversations. OpenAI is rolling out GPT-Live-1 and the lighter GPT-Live-1 mini globally for live voice applications.
What can developers use Grok 4.5 for?
Grok 4.5 is built for coding and autonomous, agentic tasks. It is available through Grok Build, Cursor and SpaceXAI’s developer console. Pricing begins at $2 per million input tokens and $6 per million output tokens, which may appeal to developers running high-volume coding or automation workflows.
How does Mistral’s Robostral Navigate help factory and warehouse robots?
Robostral Navigate enables robots to move through physical environments using a single camera. It does not require lidar, multi-camera systems or elaborate sensor arrays, and it is designed to work across robots from different suppliers. The model concentrates on navigation rather than tasks such as grasping or manipulating objects.
What would Chinese restrictions on advanced AI models mean for developers?
Restrictions could curb overseas access to advanced models from companies such as Alibaba, ByteDance and Z.ai. Western startups and researchers that rely on lower-cost Chinese open-source models for coding and agentic workloads may need to seek alternative providers. Across many pipelines, this could raise costs or deepen reliance on major US AI companies.