Cognenza · AI newshttps://cognenza.com/en/noticias/What matters in artificial intelligence, summarised and linked to the original source. No rumours: every item says where it comes from.en Fri, 16 Oct 2026 03:41:00 GMT Meta releases Muse Spark Open, an open model that can run on desktop machineshttps://cognenza.com/en/noticias/2026-10-16-manana/#n1https://cognenza.com/en/noticias/2026-10-16-manana/#n1Fri, 16 Oct 2026 03:41:00 GMTmodelsMeta has released the weights of Muse Spark Open, a multimodal model (text, image, audio and video) built for coding and long-running tasks. It can be downloaded from Hugging Face and other platforms, or used through providers such as OpenRouter or Groq. According to Meta, it is the first in its class to run locally on two high-end Mac Studios. The licence lets many developers use it for free, with limited commercial terms. Why it matters: The more powerful models you can download and run yourself, the less you depend on a single provider and the more control you have over your data. (Source: Meta AI Research) Google opens Gemini 4 Argon to developers, businesses and paying subscribershttps://cognenza.com/en/noticias/2026-10-16-manana/#n2https://cognenza.com/en/noticias/2026-10-16-manana/#n2Fri, 16 Oct 2026 03:41:00 GMTmodelsGemini 4 Argon, Google’s most advanced model, is now available in the Gemini app and in AI Mode in Search for Google AI Pro and Ultra subscribers, and via API for developers. Google pitches it at long, complex work: coding, finance, law, science and cybersecurity. Until 30 November the API has an introductory price of $2 per million input tokens and $10 output; after that it rises to $4 and $20. Why it matters: If you pay for Gemini Pro or Ultra you can try it now; and if you use it via the API, note that the price doubles from December. (Source: Google (The Keyword)) Anthropic launches Claude Haiku 5.5, its fastest and cheapest small modelhttps://cognenza.com/en/noticias/2026-10-16-manana/#n3https://cognenza.com/en/noticias/2026-10-16-manana/#n3Fri, 16 Oct 2026 03:41:00 GMTmodelsClaude Haiku 5.5 is designed for simple, repetitive, high-volume tasks: summaries, sorting messages, customer support or driving a browser. Anthropic estimates it costs about 75% less on average than Haiku 4.5. It is the first Haiku with an “effort” setting to choose between saving money and thinking more. The company has also halved the price of Sonnet 5.5 cache reads. Why it matters: Automating small tasks with AI (sorting emails, summarising documents) becomes much cheaper, opening the door to uses that were not worth it before. (Source: Anthropic) Mistral unveils Mistral Large 4, its largest model, trained in Europehttps://cognenza.com/en/noticias/2026-10-16-manana/#n4https://cognenza.com/en/noticias/2026-10-16-manana/#n4Fri, 16 Oct 2026 03:41:00 GMTmodelsFrance’s Mistral has opened a preview of Mistral Large 4 through its API and promises to publish the weights by the end of October. It is a multimodal model trained from scratch in its own data centres in Europe, on data in more than 160 languages, including every official EU language. The company highlights its performance in cybersecurity, finance and law. Why it matters: It is an open, European alternative to the big US and Chinese models, which matters for companies and public bodies that want their data to stay under European law. (Source: Mistral AI)