In partnership with
Today: the White House asks OpenAI to stagger access to its next model, Google folds computer control directly into Gemini 3.5 Flash, and Amazon, Mistral, and Krea round out a busy stretch of releases and infrastructure bets. Here are six developments worth knowing from the past two days.
A WORD FROM OUR SPONSOR
The best prompt engineers aren't typing. They're talking.
Power users figured this out early: speaking a prompt gives you 10x more context in half the time. You include the edge cases, the examples, the tone you want — because talking is fast enough that you don't skip them.
Wispr Flow captures everything you say and turns it into clean, structured text for any AI tool. Speak messy. Get polished input. Paste into ChatGPT, Claude, Cursor, or wherever you work.
89% of messages sent with zero edits. 4x faster than typing. Works system-wide on Mac, Windows, and iPhone.
Models & Products
Google bakes native computer use into Gemini 3.5 Flash. Starting June 24, Google made "computer use" a built-in capability of Gemini 3.5 Flash, letting the model operate browsers, mobile interfaces, and desktop applications by reading screenshots and issuing simulated inputs rather than running as a separate agent model. Google reported a score of 78.4 on OSWorld-Verified, a benchmark for computer-use agents, up from 65.1 for the prior Gemini 3 Flash, and said the model can now combine screen control with Google Search and Maps grounding in a single session. Google noted the leaderboard figures are self-reported and not yet independently verified. (The Decoder) (eWeek)
Mistral releases OCR 4 for enterprise document extraction. French AI company Mistral launched OCR 4 on June 23, the fourth version of its optical-character-recognition system, positioning it as an enterprise document-AI product. The model returns bounding boxes, block classification, and per-element confidence scores, supports roughly 170 languages, and can run as a single self-hosted container so documents stay on a customer's own infrastructure. Mistral priced it at $4 per 1,000 pages, halving to $2 with batch processing, and reported a top score of 85.2 on OlmOCRBench. (Mistral) (MarkTechPost)
Krea open-sources its Krea 2 image models. Krea released its Krea 2 text-to-image models in late June and published open weights under a custom community license, including a base "Raw" checkpoint and a distilled "Turbo" version. The 12-billion-parameter Turbo model generates roughly 2K-resolution images in about two seconds on consumer hardware, while Raw ships without aesthetic fine-tuning so developers can train their own LoRA adaptations without inherited style bias. The weights were posted to Hugging Face. (VentureBeat) (Hugging Face)
Policy & Access
White House asks OpenAI to stagger its GPT-5.6 release. OpenAI plans to release its next model, GPT-5.6, in stages after a request from the Trump administration, according to a June 25 report from The Information. CEO Sam Altman told staff in a memo that the government would be "approving access customer by customer" during a preview period, with a broader release expected a couple of weeks later, and added that the arrangement is "not our preferred long term model." The report cited involvement from the Office of the National Cyber Director, the Office of Science and Technology Policy, and Commerce Secretary Howard Lutnick. (TechCrunch) (Engadget)
Industry & Talent
Amazon adds $13 billion to its India AI and cloud build-out. Amazon said on June 25 it will invest an additional $13 billion in India through 2030 to expand AWS data-center capacity in Mumbai and Hyderabad, bringing its total planned India investment to about $48 billion. The company said the buildout will give startups, enterprises, and government agencies access to its custom AI chips, managed AI services, cloud technology, and developer tools. The move follows comparable India infrastructure commitments from Microsoft and Google. (CNBC) (TechCrunch)
Two Gemini researchers leave Google for Anthropic. Bloomberg reported on June 25 that Jonas Adler and Alexander Pritzel, both contributors to Google's Gemini models, plan to move to Anthropic. The departures add to a run of senior AI researcher exits from Google to rivals over the past year. Google has not publicly commented on the individual moves. (The Decoder)
That's today's edition of AI Current. Reply with what you're seeing in AI, and forward this to a colleague who would find it useful. See you tomorrow.
A WORD FROM OUR SPONSOR
Short BTC. Long ETH. Hold as Long as You Want.
Kalshi is the first CFTC-regulated perps market in the US. No expiry, no rollover, up to 5.8x leverage on BTC, ETH, SOL, XRP, and more. Trade price direction without touching the asset.
Using leverage increases risk of loss. Leverage is subject to the Firm's review and the customer's risk profile.
A WORD FROM OUR SPONSOR
Be the one who already read it. TLDR is the 5-minute tech brief that 7M+ subscribers start their morning with, curated by ex-Google and Anthropic engineers.