FrontierOpenAI released GPT-6 Sol and Luna and halved their prices against GPT-5.6's promotional ratesThe two cheaper models in the GPT-6 family came out on 22 September. Sol costs $2 for input and $10 for output per million tokens, Luna 10 and 50 cents. Per OpenAI, on its internal test Sol makes about half as many factual mistakes as its predecessor.FrontierGrok 4.7 is a bigger model at the old price - and it ships with a table that shows where it falls behindOn 21 September SpaceXAI released Grok 4.7: a new, larger base model, trained for longer, weighted toward tasks that take hours, at the same price as Grok 4.6. In the company's own table Fable 5.1 stays ahead on CursorBench and in the terminal, but it costs several times more.FrontierAlibaba opened Qwen-Image-2.1 for download, but its licence allows research and evaluation onlyQwen's new image model has 7 billion parameters in its generation component, makes transparent images and takes up to 10 reference pictures. The weights are free to download under the Qwen Research License, which forbids commercial use without a separate agreement. Alongside it Alibaba released Qwen3.8-Omni-Flash, but announced that one through its cloud, without open weights.FrontierOpenAI turned GPT-6 Astra into a lawyer: 54 instead of 38.7 per cent correct answers on a legal testAstra for Law combines OpenAI's newest model with an index of US law spanning over 230 million URLs and instructions for legal analysis. It goes first to selected law firms, and Harvey and Legora will build on it.FrontierFirefox's assistant is moving to Mistral's models in France and North AmericaSmart Window, Mozilla's beta browsing assistant, now runs on Mistral models for users in France and North America. By default conversations do not stay on Mozilla's servers. Mistral has agreed to zero data retention.FrontierSalesforce built its own model on the open Nemotron and says it used not a single byte of customer dataKoa is Salesforce's first reasoning model, post-trained from NVIDIA Nemotron 3 Super on synthetic data from nearly three decades of CRM. Per NVIDIA, Salesforce holds the weights and the model runs only in its own infrastructure.FrontierDeepSeek released V4.1-Flash and retired its own flagship: requests to V4-Pro now go to the smaller modelOn 10 September DeepSeek released V4.1-Flash, the smallest model of a new architecture family, with 552 billion parameters and only 8 billion active when reading input. From 14 September all requests to V4-Pro are routed to it at its prices. The weights are under MIT.FrontierMistral raised 3 billion euros and the lead investor is SamsungMistral announced a 3 billion euro round at a valuation above 21 billion - in the company's words the largest ever completed by a European technology firm. Samsung Electronics leads.
80 stories