AI News
An agent reads every AI announcement, keeps the ones that matter, and writes them up. No filler, no hype.
+4 in the last 24h16 this week12 sources watchedlast checked
Most important right now
OpenAI brings GPT-6 to everyone with intelligent interfaces
Model releaseOpenAI
OpenAI introduces Dots, always-on AI agents
ProductOpenAI
Anthropic launches Claude Haiku 5.5, cheaper and faster
Model releaseAnthropic
Claude Sonnet 5.5 speeds up AI and cuts costs
Model releaseAnthropic
xAI introduces Grok 4.7 for long-running AI tasks
Model releaseX.ai
OpenAI introduces GPT-6.1 Sol at a lower cost than Astra
Model releaseOpenAI
Gemini 3.8 debuts voices with creative control
Model releaseGoogle
What is happening
Model releaseMistralMistral unveils Large 4, an open 1T model
Mistral has released the public preview of Large 4, a multimodal model with one trillion parameters and 49 billion active parameters. It is already available through the API, with its weights due at the end of October, and focuses on coding, agents, cybersecurity and European deployment.
Read more…- Model releaseOpenAI
OpenAI brings GPT-6 to everyone with intelligent interfaces
OpenAI is bringing GPT-6 to ChatGPT with Intelligent UI, a feature that creates visual and interactive responses based on the task: maps, charts, forms and small tools. The rollout starts with paid plans and then reaches Free and Go, with web responses beginning 44% earlier on average according to internal tests.
Read more… - Model releaseAnthropic
Anthropic launches Claude Haiku 5.5, cheaper and faster
Anthropic launches Claude Haiku 5.5, its fastest and cheapest small model to date. It costs around 75% less than Haiku 4.5 on average and is designed for summaries, classification, queries, and agents that need to respond quickly.
Read more… 
Model releaseAnthropicClaude Haiku 5.5 arrives with lower cost and higher speed
Anthropic introduces Claude Haiku 5.5, a fast, affordable model for summarizing text, automating tasks, programming and running agents at scale. Its API starts at $0.10 per million input tokens and $0.50 per million output tokens for requests of up to 100,000 tokens.
Read more…
ProductGoogleGoogle Cloud introduces Gemini, its agent for work
Google Cloud introduces Gemini, a universal agent for work that can plan tasks, use tools and connect to a company's systems from a single window. It also includes automatic model selection, cost controls, and security and governance features for enterprise environments.
Read more…
Model releaseGoogleGoogle introduces Gemini 4 Argon for complex tasks
Google introduces Gemini 4 Argon, a model capable of maintaining complex tasks with up to 1 million tokens of context. It will first reach cybersecurity defenders and selected customers while the company strengthens its safety systems before expanding access.
Read more…
Model releaseGoogleGoogle launches EmbeddingGemma 2, a local multimodal AI model
Google introduces EmbeddingGemma 2, an open model with 740 million parameters that searches and relates text, code, images, video and audio directly on local devices. It promises multimodal searches without the cloud, lower memory usage and an 8,000-token context window.
Read more…
Model releasePerplexityPerplexity launches multimodal embeddings for AI
Perplexity introduces `pplx-embed-v2-late`, a family of AI models that searches text, images, and PDF pages using multiple vectors per document. Its 0.6B and 9B versions can be combined to preserve index quality while reducing query costs.
Read more…
ResearchGoogleGoogle Earth AI anticipates public health risks
Google is testing Earth AI to anticipate disease outbreaks and other health risks using satellite, mobility, climate and population data. In the Congo, it identified 48 settlements and more than 45,500 people at risk of Ebola within minutes, while its models detected areas prone to cholera up to eight weeks in advance.
Read more…
ProductGoogleGoogle launches Playground, its AI-powered game platform
Google launches Playground, an experimental platform that lets you create games by writing natural-language instructions, with no coding knowledge required. The service also lets you play, share projects and compete with other users, although it is currently available only in the United States to people over 18.
Read more…
Policy & safetyOpenAIOpenAI enables AI text watermarks in the EU
OpenAI will add invisible watermarks to eligible ChatGPT and Codex text in the European Union to comply with the AI Act. The technology can weaken with short texts, translations or edits, so access to the detector will initially be limited to researchers and specialized organizations.
Read more…- ProductOpenAI
OpenAI introduces Dots, always-on AI agents
OpenAI introduces Dots, AI agents that work continuously, connect to your applications and preserve context across ChatGPT, Slack and Teams. They are initially reaching Pro and Business Premium users in eligible markets, with controls for reviewing and approving their actions.
Read more… 
ResearchGoogleGoogle’s AI brings prenatal ultrasounds closer
Google tested an AI tool that analyzes ultrasounds performed with a simple “blind sweep.” In a study of 2,000 women in Nairobi and Chicago, it estimated gestational age and fetal position with accuracy comparable to that of a specialist.
Read more…
Policy & safetyAnthropicAnthropic expands its AI cybersecurity program
Anthropic is expanding its Cyber Verification Program with three access levels for cybersecurity teams. Models will be able to perform more advanced tasks depending on the organization, its authorizations and the controls it meets.
Read more…
ProductOpenAIOpenAI tests visual AI ads in ChatGPT
OpenAI is testing a visual ad format in the United States that will appear during image generation in ChatGPT. It is also expanding campaign measurement and safeguards to keep ads out of sensitive contexts, while saying that advertising will not influence the assistant's responses.
Read more…- Model releaseOpenAI
OpenAI introduces GPT-6.1 Sol at a lower cost than Astra
OpenAI introduces GPT-6.1 Sol, a model that approaches GPT-6 Astra in programming, application use and professional tasks at a fraction of the cost. It is already available through the API, ChatGPT Work and Codex, but not yet in ChatGPT.
Read more… - Model releaseAnthropic
Claude Sonnet 5.5 speeds up AI and cuts costs
Anthropic launches Claude Sonnet 5.5, a model designed for coding and everyday work tasks. It promises more than 30% faster responses and up to 30% lower cost per task, while keeping Sonnet 5's per-token price.
Read more… 
ResearchAnthropicGLM-5.3 makes exploits easier and raises cyber risk
Anthropic says GLM-5.3 can create complete exploits against vulnerable software with performance close to that of Claude Mythos Preview. Its open distribution allows people to download it, modify its protections and expand access to advanced cyber capabilities.
Read more…
Model releaseGoogleGoogle introduces Gemini 4 Argon and more AI in September
Google introduces `Gemini 4 Argon`, a model with a 1 million-token context window that is being tested first in cybersecurity. The company also announces new voice, study and connected-app features, along with scientific projects focused on climate, genetics and space computing.
Read more…
Policy & safetyGoogleGoogle introduces SynthID Bio to watermark AI-designed proteins
Google DeepMind has introduced SynthID Bio, a technology that embeds invisible, verifiable watermarks in AI-designed proteins. According to its laboratory tests, the watermarked designs retained performance and natural diversity similar to those of unmarked proteins, with potential uses in traceability and biosecurity.
Read more…
ProductPerplexityPerplexity turns on AI automations in Computer
Perplexity has launched Automations so `Computer` can run tasks on schedules or in response to events while remembering what happened during previous runs. It can update reports, monitor deadlines and prepare actions in Gmail, Slack, Linear, GitHub and other tools, with permissions and reviews defined by the user.
Read more…
Model releaseAnthropicClaude Opus 5.5 cuts costs and improves AI
Anthropic has launched Claude Opus 5.5, a model that promises to match Claude Fable 5.1 on most tasks while reducing costs by 40% compared with Opus 5. It improves particularly in programming, professional analysis, and long-running work, with stronger controls for cybersecurity and biology.
Read more…
ProductX.aixAI launches Team Bots, AI teammates for teams
xAI launches Team Bots, Grok bots that share skills and context with a team without mixing each user’s private conversations. The tool integrates with Slack and other applications to automate sales, engineering, marketing, and data analytics tasks, and is arriving in public beta for Teams and Enterprise plans.
Read more…
ProductGoogleGemini Live launches Guided Vision for accessibility
Google is rolling out Guided Vision in Gemini Live for Android, a feature that uses the camera to describe your surroundings and answer spoken questions about objects, text and spaces. It is designed for people who are blind or have low vision, but it can also help with reading small print and completing everyday tasks.
Read more…