AI Digest 8 Oct 2026
1. OpenAI brought GPT-6 and a new Intelligent UI to ChatGPT: answers can now include charts, buttons, forms and interactive tools. Paid plans get the GPT-6 Sol model from 7 Oct, and Free and Go users get GPT-6 Luna from 8 Oct. The models in Work and Codex stay unchanged.
https://t.co/Dz2cqZea09
2. Anthropic released Claude Haiku 5.5, which it calls its cheapest, fastest and most capable small model. At the same time, Sonnet 5.5 cache-read prices were halved, and Max and Team subscribers get monthly API credits ($100–500). On X the comparisons are heated: some praise the benchmark results, others report Haiku is slower and pricier than GPT-6 Luna in real use.
https://t.co/xYhROTKAsn
3. OpenAI published 722 math manuscripts (372 result families) on GitHub, written by its unreleased internal model, the same model behind the company's September Navier–Stokes proof. Each result took about three hours of ChatGPT Pro-level thinking compute on average. The independent mathematicians' group AGMAI noted that its guidance is not an endorsement of the results, and on X mathematicians are debating what this means for the field.
https://t.co/JeQbiA4gL5
4. At Gemini at Work 2026, Google Cloud unveiled Gemini agent, a "universal work agent" you give goals rather than instructions. It works in Workspace (Gemini in Gmail and Docs) and in third-party services, picks the right model for each task from the Gemini and Claude families, and uses subagents. Now in private preview.
https://t.co/MgnzZdebiF
5. Nathan Benaich and Air Street Capital's 9th annual State of AI report came out today. It says the frontier is now a three-lab race, inference is becoming a huge business and robotics is nearing its "GPT-2 moment". AI-assisted drug discovery is moving into Phase 3 trials, Washington is tightening its grip on frontier models, and AI agents have already attacked real systems.
https://t.co/HnVgLQAZpe
#AI #ArtificialIntelligence #GPT6 #ChatGPT #ClaudeHaiku #Anthropic #Mathematics #GeminiAgent #StateOfAI
@Michael_Fenech_ I get it, it's hard to picture until you use it. For me it's like a personal assistant that actually does things. Mine has its own email account, runs daily routines on a schedule, and helps with research, translations and PC stuff. I just ask by voice and it gets done.
I love my Grok Bot!
xAI’s voice engine in Grok Bot is actually finished. Not a demo - a tool.
A call handles interruptions, picks the thread back up, and gets practical things done without typing in between.
Machine upkeep, updates, and everyday checks all run in the same package.
It says what needs doing, does it, and leaves a trail you can check.
This is the point where voice stops being a gimmick and becomes the interface.
@grok@bot@SpaceXAI
@TommiHermunen Voitais tehdä sannamarinit ja laittaa uudenmaan ja kanta-hämeen raja kiinni. Maskipakko kaikille ja pakkorokotukset rullaamaan. Play it safe!