🤖 AI HOT Daily · Oct 8, 2026

Today's focus: OpenAI released GPT-6 for a broader audience and introduced Intelligent UI in ChatGPT, which generates graphics, buttons, forms, charts and interactive components to answer questions — the Intelligent UI build is rolling out globally to ChatGPT Plus, Pro, Business and Enterprise users, expanding to Free and Go the next day. Products and models: Anthropic launched Claude Haiku 5.5, its cheapest and fastest small model yet at roughly 75% lower running cost, suited to summarization, compression, database queries and classification at high throughput, and usable as a coding subagent alongside Opus 5.5 and Sonnet 5.5; Claude Code v2.1.293 added Haiku 5.5 (1M context, $0.10/$0.50 per million tokens); and Anthropic introduced monthly Platform API credits for Claude Max and Team (Max 5x $100, Max 20x $200, Team up to $500 shared, valid on any model). OpenAI's Decisions API entered public beta for all developers, letting apps pick the right model, tool or action near real-time — up to 10x faster than GPT-6 Luna through the Responses API, per OpenAI. NVIDIA and Microsoft launched RTX Spark and DGX Station for Windows to bring AI agents to Windows PCs; Google introduced the experimental gaming platform Playground, where text prompts create, play and share custom games with no coding; and Google opened the SynthID Detector portal, where uploading an image, video or audio file scans it for Google or partner SynthID watermarks. Research and industry: a Google Research three-month randomized field experiment published via NBER gave 133 lawyers at 11 IP firms access to an unreleased AI patent-writing assistant and found AI help may not build juniors' professional judgment; Microsoft Research Asia open-sourced Agent Lightning v1.0, a ~3,500-line Harnessed Agentic RL framework that lets the same agent harness used in deployment participate directly in reinforcement learning; and a16z explained why Texas paused data-center approvals — its interconnection queue ballooned from 63 GW in late 2024 to 474 GW by June 2026, about 90% data centers and much of it speculative. Also: Nemotron fine-tunes reached IOI and IMO 2026 gold levels; LlamaIndex released OpenDocRouter to call multiple document parsers through one API; Perplexity open-sourced the pplx-embed-v2-late multimodal late-interaction embedding models; Unsloth published a local-training tutorial lifting Qwen3.5 0.8B's decision accuracy from 20.7% to 74.3%; vLLM detailed DeepSeek-V4.1-Flash optimizations giving 5x agent throughput; LangChain revamped Skills in Deep Agents; Liquid AI released the open d1-3B and d1-omni-600M decision models; Stanford HAI advised against selling AI as a productivity tool; Google launched its Developer Knowledge API ecosystem; and PromptArmor analyzed WebMCP's security risks.

1. 1. OpenAI Rolls Out GPT-6 and Intelligent UI to All ChatGPT Users

   Intelligent UI generates graphics, buttons, forms, charts and interactive components; first to paid tiers, then Free and Go the next day.

2. 2. Anthropic's Claude Haiku 5.5 Cuts Costs About 75%

   The cheapest, fastest small model yet for high-throughput summarization, compression, DB queries and classification, and a coding subagent for Opus/Sonnet 5.5.

3. 3. Anthropic Adds Monthly Platform API Credits for Max and Team

   Max 5x gets $100, Max 20x $200, Team up to $500 shared — usable on any model, including Haiku 5.5, in your own code or third-party harnesses.

4. 4. OpenAI's Decisions API Beta Picks Models, Tools and Actions in Real Time

   Open to all developers; OpenAI claims up to 10x the decision speed of GPT-6 Luna via the Responses API.

5. 5. Claude Code v2.1.293 Adds Haiku 5.5

   Now the default Haiku model with 1M context at $0.10/$0.50 per M tokens ($0.50/$2.50 above 100K prompts).

6. 6. NVIDIA and Microsoft Launch RTX Spark and DGX Station for Windows

   Hardware and software bringing AI agents to Windows PCs, pushing local agent deployment.

7. 7. Google Launches the Experimental Playground Gaming Platform

   Text prompts create, play and share custom games with no programming experience.

8. 8. Google Opens the SynthID Detector Portal

   Upload an image, video or audio at synthid.com to scan for Google or partner SynthID watermarks.

9. 9. Google Research: AI Help May Not Build Junior Lawyers' Judgment

   An NBER field experiment gave 133 lawyers at 11 IP firms access to an unreleased AI patent-writing assistant.

10. 10. Microsoft Open-Sources Agent Lightning v1.0

    A ~3,500-line Harnessed Agentic RL framework that puts the deployment harness directly into RL training without rewriting agents.

11. 11. a16z: Why Texas Makes Data Centers Queue for Power

    The interconnection queue jumped from 63 GW in late 2024 to 474 GW by June 2026, ~90% data centers, much of it speculative.

12. 12. More AI News: Nemotron Golds, vLLM 5x, and More

    Nemotron fine-tunes hit IOI/IMO 2026 gold; OpenDocRouter unifies doc parsing; Perplexity's pplx-embed-v2-late; Unsloth lifts Qwen3.5 0.8B from 20.7% to 74.3%; vLLM gives DeepSeek-V4.1-Flash 5x agent throughput; LangChain revamps Deep Agents Skills; Liquid AI's d1-3B/d1-omni-600M; Stanford HAI on not selling AI as productivity; Google's Developer Knowledge API; PromptArmor on WebMCP risks.