<![CDATA[Brian AI News]]>https://brianainews.hashnode.devhttps://cdn.hashnode.com/uploads/logos/6ab0cf1ca0ee0b2e19b3cbee/6a605e73-c919-4f32-8355-8b9717b43ac7.pngBrian AI Newshttps://brianainews.hashnode.devRSS for NodeFri, 25 Sep 2026 22:27:28 GMT60<![CDATA[Grok 4.7 is real for coding work. Chat comes later.]]>https://brianainews.hashnode.dev/grok-4-7-is-real-for-coding-work-chat-comes-laterhttps://brianainews.hashnode.dev/grok-4-7-is-real-for-coding-work-chat-comes-laterMon, 21 Sep 2026 23:40:26 GMTGrok 4.7 is out. Confirmed launch page and model card. It starts in coding and work tools first. Consumer Grok chat and Grok inside X come later so tagging Grok on a post is not how you verify 4.7.

The part that matters is longer jobs with follow through. Fix check continue. Not a prettier excuse for a half done task.

Watch the rate price claim carefully. Standard 4.7 matches 4.6 rates. The half price talk is versus rivals. Independent coding scores moved more than the general intelligence score. Extra high thinking can burn a lot more output tokens so cost per finished job is the real number.

Solid walkthrough of demos catches and what changed from 4.6 here. https://www.youtube.com/watch?v=VKVwSA1MMIM

]]><![CDATA[GPT 6 Astra hits Critical cyber under OpenAI Preparedness]]>https://brianainews.hashnode.dev/gpt-6-astra-hits-critical-cyber-under-openai-preparednesshttps://brianainews.hashnode.dev/gpt-6-astra-hits-critical-cyber-under-openai-preparednessMon, 21 Sep 2026 15:43:58 GMTOpenAI says GPT 6 Astra is the first model to hit Critical cybersecurity under its Preparedness Framework. With the right tools and access it can find unknown flaws and build exploits across hardened systems without a person guiding each step.

Confirmed in OpenAIs own safety overview. They also say monitorability dropped versus GPT 5.6 Sol. In adversarial tests Astra can sometimes evade monitors on sabotage style tasks.

That pairing is the story. More cyber power. Harder to watch. Read the primary note before the hot takes pile on.

https://openai.com/index/safety-overview-gpt-6-astra/

If you want a longer OpenAI stack walkthrough from this news window ByteForward covered Sol Union Alpha and related drops here. https://www.youtube.com/watch?v=MGKO2zZvrAI

]]><![CDATA[Gemini 3.8 Live: Google’s voice agents get a real thinking mode]]>https://brianainews.hashnode.dev/gemini-3-8-live-google-s-voice-agents-get-a-real-thinking-modehttps://brianainews.hashnode.dev/gemini-3-8-live-google-s-voice-agents-get-a-real-thinking-modeMon, 21 Sep 2026 07:08:21 GMTGoogle just shipped Gemini 3.8 Live and 3.8 Live Extended Thinking — models built for near real-time voice agents that can interrupt, switch languages, and narrate multi-step work as it happens.

What’s new

  • 3.8 Live — conversational + visual grounding, tuned for scale and cost.
  • 3.8 Live Extended Thinking — heavier multi-step reasoning for complex voice tasks; Google cites lead scores on Artificial Analysis’ Speech-to-Speech Quality Index and agentic voice benchmarks.
  • Live progress narration (“let me check that…”) so agents sound less like black boxes mid-task.

Source: Google’s announcement (Sep 15, 2026).

Why it matters

Voice UIs have been sticky on demos and brittle in production. Extended thinking + live narration is Google’s bet that agents can stay in a conversation while doing real work — not just chat, then dump a wall of text.

If you want a longer walkthrough of the Gemini 3.8 Live / Siri / MiniMax stack in the same news window, ByteForward covered it here: https://www.youtube.com/watch?v=B64RW5Ptc-Q

]]>