Skip to content
Source · Daily Brief

AI Daily Brief — 30 July 2026

Google DeepMind and OpenAI headlined a busy Thursday, pairing a whole-body Gemini Robotics 2.0 update with a price-performance GPT-5.6 release. Microsoft Research, AWS and NVIDIA filled out the agent-and-infrastructure layer while arXiv and Hugging Face carried a heavy wave of papers on self-evolving agents, distillation and robotics.

Top stories

  • OpenAI releases GPT-5.6. The company frames the model as a push along the price-performance frontier. via OpenAI
  • Gemini Robotics ER 2 arrives. Google DeepMind’s release adds video understanding, task orchestration and multi-robot collaboration. via Google DeepMind
  • Gemini Robotics 2.0 can control a robot’s entire body. Coverage highlights improved dexterity and safety across full-body motion. via The Verge – AI
  • Kimi K3 lands on AWS. A new deployment guide walks through serving the model on Amazon infrastructure. via AWS Machine Learning Blog
  • Microsoft Research unveils Echoverse. The project builds deep, evolving environments for training computer-use agents. via Microsoft Research
  • New MCP spec targets enterprise scale. A stateless redesign aims to make the protocol easier to run at large deployments. via Ars Technica – AI
  • A fundamental flaw leaves LLMs vulnerable. Researchers describe a structural weakness that opens models to attack. via MIT Technology Review – AI
  • GitHub Copilot ships its July updates. The VS Code release rounds up a month of changes to the coding assistant. via GitHub Copilot changelog

Who shipped

OpenAI released GPT-5.6, emphasising cost and throughput. Google DeepMind pushed Gemini Robotics ER 2 and a Gemini Robotics 2.0 update centred on whole-body control, dexterity and safety. Microsoft Research published Echoverse and EvoLib, both aimed at computer-use agents and evolving knowledge. AWS added a Kimi K3 deployment path plus explicit prompt caching for GPT-5.6 on Bedrock. NVIDIA shipped developer material across nvmath-python, Exemplar Cloud and secure-agent guidance, while GitHub rolled out Copilot updates for Visual Studio and VS Code and Sakana AI ran a product-development interview.

Open-source pulse

llama.cpp shipped a rapid run of tagged builds today, from b10184 through b10200. The research feed leaned open, with Hugging Face Daily Papers and arXiv carrying work like SkillRise, DecoEvo, CADENCE, Weak-to-Strong On-Policy Distillation and TurboVLA, a real-time vision-language-action model running at 32 Hz on a single RTX 4090. Open-weight momentum showed up mostly through tooling, such as the AWS guide for serving Kimi K3, rather than a fresh headline model drop.

Quiet corners

Anthropic was silent with no Claude news, and Meta had no Llama update. xAI, Mistral, DeepSeek and Qwen stayed quiet as well; the only China-lab presence was Moonshot‘s Kimi appearing through an AWS deployment guide rather than a direct release.

Money, infra & hardware

On the consumer side, Tim Cook hinted at an iCloud Plus tier aimed at AI power users. via The Verge – AI On the datacenter side, NVIDIA laid out lessons from its Exemplar Cloud for unlocking full performance on AI infrastructure. via NVIDIA Developer Friend also relaunched its AI pendant with an added speaker at twice the price.

By the numbers

  • 129 stories across 19 sources
  • Most-mentioned model: Gemini
  • Most-mentioned lab: Google DeepMind
  • Notable absences: Anthropic, Meta

Compiled by AI Feed’s editor from all 129 headlines published on 30 July 2026.