As of June 29, the strongest models from OpenAI and Anthropic have both entered Washington's new "frontier-model review" gate: OpenAI previewed the three-tier GPT-5.6 (Sol/Terra/Luna) on June 26, with access government-gated; Anthropic's Mythos 5 was also partially cleared on June 27, released first to critical-infrastructure defenders. Neither is yet publicly available.
On June 28, Musk announced that xAI's Grok 4.5 has entered internal private beta at Tesla and SpaceX. It's built on the new V9 base model (1.5 trillion parameters), with supplementary training using Cursor's data. Musk says early evaluations approach or even surpass Claude Opus—a claim currently coming from him alone.
On June 27, GPT-4.5 was retired from ChatGPT as its 30-day sunset period ended. This was already announced in the May 28 release notes, and after retirement users' existing conversations auto-migrate to the corresponding GPT-5.5 model to continue.
On June 26, CNBC reported that companies are no longer blindly "use as much as you can" pouring tokens, and are starting to calculate returns. Uber set monthly spending tiers for some AI tools, and startup Lindy simply switched all its traffic from Claude to the cheaper DeepSeek.
On June 25, at its Investor Day, Qualcomm unveiled a data-center roadmap, introducing the Dragonfly C1000 CPU, the AI300 inference accelerator, and high-bandwidth computing solutions aimed at agentic AI workloads. Meta will adopt it after mass production in 2028, and Qualcomm set a target of $15 billion in data-center revenue by 2029.
On June 24, TechCrunch reported that two core figures in Gemini's development, Jonas Adler and Alexander Pritzel, left Google for Anthropic; about a week earlier, Noam Shazeer, one of the Transformer authors, also left—for OpenAI. This wave of departures coincides with the two companies preparing for IPOs and having equity in hand to poach talent.
On June 23, according to CNN, the US and its "Five Eyes" intelligence partners issued a rare joint warning: as AI attack capabilities advance rapidly, malicious actors may breach government and corporate network defenses within months. It's the first time these countries' intelligence agencies have given such an urgent timeline for the AI cyber threat.
On June 22, under its Daybreak security program, OpenAI released three things at once: the full version of GPT-5.5-Cyber, open only to vetted security institutions; a Codex Security plugin embedded in the dev workflow; and Patch the Planet, dedicated to finding and fixing bugs in open source projects.
In late June, France's Mistral shipped a run of updates: Medium 3.5 entered Le Chat and Vibe, powering a remote coding agent and Work mode, with API pricing of $1.5/M input and $7.5/M output; alongside it, OCR 4 launched, supporting 170 languages and self-hostable in a single container.
On June 19, John Jumper—lead author of AlphaFold and 2024 Nobel laureate in Chemistry—announced he's leaving Google DeepMind, where he spent nearly nine years, for Anthropic. What exactly he'll do at his new employer hasn't been announced, but the move is seen as a signal of the frontier labs poaching talent for "doing science with large models."
On June 19, multiple outlets reported that OpenAI acquired the Python-tooling company Astral—the maker of the fast package manager uv and the linter/formatter ruff—aiming to fold these tools into its coding product Codex. Deal terms haven't been disclosed by either side.
On June 18, the US Federal Energy Regulatory Commission issued "show cause" orders to all six regional grid operators, requiring them within 60 days to justify their existing large-user interconnection rules or produce reform proposals, so that large loads like AI data centers can connect to the grid faster.
Showing 12 of 12 stories
