Daily AI News

xAI releases Grok 4.5, trained on tens of thousands of Nvidia GB300 GPUs

xAI has launched Grok 4.5, a new model trained on massive GPU clusters that trails Fable 5 and GPT-5.5 on coding benchmarks but costs dramatically less at $2 per million input tokens — 4.2x fewer tokens than Opus 4.8. EU availability expected mid-July.

Read source →

Meta enters AI coding battle with Muse Spark 1.1

Meta launched Muse Spark 1.1, positioning it for large agentic workloads, bug fixing, and large code migrations — targeting enterprise automation needs in the increasingly competitive AI coding assistant market.

Read source →

NVIDIA releases Nemotron Labs 3 Puzzle 75B: compressed hybrid MoE delivering 2.03x server throughput

NVIDIA's new compressed variant of Nemotron-3-Super uses iterative puzzle compression to drop from 120.7B to 75.3B total parameters while achieving 2.03x server throughput on a single H100, maintaining performance through knowledge distillation recovery.

Read source →

Databricks adopts Chinese open-source model GLM 5.2 as default coding engine after matching Opus at lower cost

Databricks benchmarked GLM 5.2 against Anthropic's Opus 4.8 on its multi-million-line codebase, finding it matched performance at $1.28 per task vs $1.94, and will roll it out as daily coding workhorse — highlighting open models closing the gap with proprietary leaders.

Read source →

OpenAI sunsets Atlas AI browser, moves agentic features to desktop app and Chrome extension

OpenAI is shutting down its AI-powered browser Atlas after less than a year, but integrating its agentic browsing capabilities into the ChatGPT desktop app and a Chrome extension, signaling a pivot in distribution strategy.

Read source →

Google launches "created or edited with AI" labels for ads across Search, Discover, and YouTube

Google added transparency labels in My Ad Center showing when ads were made or edited with AI, covering Search, Discover, and YouTube — a platform-wide move toward AI disclosure in advertising.

Read source →

AI agent startup Lyzr lets its own agent run $100M fundraise

Enterprise AI agent company Lyzr used its own AI agent to conduct a $100 million funding round, demonstrating the agent's capability to handle complex, high-stakes business processes autonomously.

Read source →

Meta's new AI chips to begin production in September with modular design

Meta's custom AI accelerators enter production this fall using a modular architecture designed to adapt to rapidly evolving AI workloads, reducing dependence on Nvidia for its massive infrastructure needs.

Read source →

OpenAI system beats every human at AtCoder World Tour Finals 2026

An OpenAI system solved all five problems in the Algorithm Division at the AtCoder World Tour Finals, including two exceptionally difficult problems, crushing all human competitors in an exhibition match.

Read source →

New York Times alleges OpenAI hid evidence in ChatGPT copyright trial

News publishers escalated their lawsuit against OpenAI with a motion for sanctions, alleging the company concealed tools and datasets that could identify copyrighted journalism in ChatGPT outputs.

Read source →