AI news · Topic
New AI models and features
Every major AI model release and feature update, summarised in plain words: what changed, what it costs and whether it matters for everyday users. Covers OpenAI, Anthropic, Google, Mistral and more.
- Mistral lets anyone download its biggest AI model, if they have the hardware (Mistral, ). On 29 October, Mistral released the weights of Mistral Large 4, its largest model so far. Weights are the files that make a model work, so companies can run it on their own machines.
- Google's Gemini 4 Carbon goes to security experts first (Google, ). Google announced Gemini 4 Carbon on 28 October, a model built for hard coding and cybersecurity work. It starts with a small group of security experts while Google keeps testing its safeguards.
- OpenAI's GPT-6.1 Terra uses a computer faster, at half the price (OpenAI, ). GPT-6.1 Terra arrived on 26 October. It is built for using a computer, reading images, coding and everyday work, and it costs developers half as much as the models OpenAI compares it with.
- Claude Haiku 5.5 is much cheaper, but read the small print (Anthropic, ). Anthropic released Claude Haiku 5.5 on 7 October. For shorter requests it costs 90% less than Haiku 4.5, and Anthropic puts the average saving at about 75%.
- ChatGPT gets “dots”: AI helpers that keep working while you're away (OpenAI, ). Dots are AI helpers in ChatGPT that keep working on a goal over days. Each one has its own computer in the cloud, learns from your feedback, and can connect to thousands of apps. They are starting with paid business plans.
- Claude Sonnet 5.5 is out: same price, and about 30% faster (Anthropic, ). Sonnet 5.5, the middle model in Claude's range, arrived six days after Opus 5.5. The price stays the same as Sonnet 5, while Anthropic says it runs over 30% faster and costs up to 30% less per task.
- Anthropic releases Claude Opus 5.5: faster, and 40% cheaper to run (Anthropic, ). Opus 5.5 is the first model in Claude's new 5.5 family. Anthropic says it does most work as well as its top model, runs over 30% faster, and costs 40% less to use than Opus 5.
- OpenAI launches GPT-6 Sol and Luna at about half the old price (OpenAI, ). GPT-6 Sol is OpenAI's model for everyday work, coding and using a computer. Luna is a very cheap, fast version. Both cost about half as much as the GPT-5.6 models they replace.
- Google's Gemini 3.8 Flash scores better at the same price, until January (Google, ). Gemini 3.8 Flash is Google's fast, low-cost everyday model. It does better than the version before it at the same price. Google says the price will double on 1 January 2027.