Numerama’s weekly roundup highlights OpenAI’s 772 mathematical manuscripts, which impressed scientists, and Mistral AI’s return with its first model in months. It also covers Orange’s first home cinema set-top box.
TypeSafe, maker of the Jev AI model, is valued at $7.5 billion only weeks after launch. The company claims Jev runs much faster and uses far fewer tokens than large language models.
Asana says its browser agent became 76 times cheaper and 5 times faster in tests. The company used OpenAI models in Codex to get there and plans to offer customers more capable models.
Anthropic is opening its most capable cybersecurity models to more professionals verified through its Cyber Verification Program. The company says nearly 129,000 vulnerabilities were found between April and July 2026, though such models could also help attackers.
A Hugging Face blog post looks at building your own AI model when no suitable one exists.
A Hugging Face post introduces d1, open multimodal decision models designed to run on edge devices.
Falcon ASR, a speech recognition model, has been introduced on the Hugging Face blog.
As of Oct. 6, 2026, the price of one million output tokens ranges from $0.50 (GPT-6 Luna) to $50 (GPT-6 Astra and Claude Fable 5.1) across the 18 offers compared by ActuIA.
GPT-6 is rolling out worldwide in ChatGPT together with Intelligent UI. Responses are faster and can include visuals and interactive elements users can work with directly.
Google DeepMind has released EmbeddingGemma 2, an open and lightweight model that produces multimodal embeddings.
On Oct. 6, Mistral opened a preview of Mistral Large 4, a 1.05-trillion-parameter model it calls the most capable open model outside China. Independent measurements show it catching up, with cybersecurity pitched as a selling point against Chinese open models.
Google has updated its documentation to say that from Oct. 9, users without a subscription will lose access to Gemini Flash and Pro. The free tier will be significantly restricted.
On Oct. 3, German company Aleph Alpha released Kolibri 1, a mixture-of-experts language model whose full weights can be downloaded from Hugging Face. ActuIA examines what its Apache 2.0 license allows.
OpenAI has published a guide for startups on choosing between GPT-6 models. It covers tuning reasoning effort, writing prompts and skills, coordinating tools and preparing workflows for production.
Google has published a roundup of the AI updates it announced in September 2026.
GPT-6 Astra Ultrafast, which runs on NVIDIA Blackwell GPUs, is now available in the OpenAI API and to eligible ChatGPT Work and Codex users. It generates tokens up to 8 times faster than Astra’s standard mode.
Google DeepMind has announced Gemini 4 Argon, which it presents as the start of a new era for its frontier models.
OpenAI has summarized more than 20 announcements from its DevDay 2026 developer conference. They cover GPT-6 Astra, ChatGPT, Codex, APIs, security and new developer tools.
OpenAI has released GPT-6.1 Sol, which it says comes close to Astra’s level on coding, computer use and professional work. Its API costs one-fifth of Astra’s standard price for input and output tokens.
A Hugging Face post presents Holo4, designed to power general-purpose agents that operate computers.
Accounting AI firm Basis says GPT-6 Astra finished a 50-tab tax workbook twice as fast as GPT-5.6 Sol. Its better grasp of user intent also makes the company more confident using it in real conditions.
Google DeepMind has introduced a Live Avatar feature for Gemini 3.8 Live.
Google DeepMind has announced text-to-speech capabilities for Gemini 3.8.
OpenAI has improved prompt caching for GPT-6 with higher hit rates, new diagnostics, explicit breakpoints and extra controls. The changes are meant to lower latency and cost for developers.
OpenAI has released two new models, GPT-6 Sol and GPT-6 Luna. They offer different trade-offs between capability and cost for everyday work.
Google DeepMind has announced Gemini 3.8 Live along with a 3.8 Live Extended Thinking variant.