Washington restored global access to Anthropic's Fable 5 and Mythos 5 after 19 days offline, the same week Google shipped DiffusionGemma and Etched raised $800M to challenge Nvidia.
The startup raised $800M and locked more than $1B in signed customer contracts, marking the strongest challenge yet to Nvidia with purpose-built transformer accelerators.
The Commerce Department lifted export licensing on Fable 5 and Mythos 5, restoring public access on July 1 after a 19-day outage tied to the earlier jailbreak crisis.
DiffusionGemma applies diffusion-style parallel token generation to autocomplete and code infilling, while Gemini 3.5 Flash hit general availability July 1 for large-scale agentic work.
Google launched DiffusionGemma, an experimental mixture-of-experts model using diffusion-style parallel token generation for local autocomplete, inline editing, and code infilling, running several times faster than existing Gemma models. Gemini 3.5 Flash reached general availability on July 1, tuned for large-scale...
Continue reading arrow_outwardGoogle unveiled DiffusionGemma alongside Gemini 3.5 Flash, pushing diffusion-based text generation and cheaper inference into the mainstream.
US and Iranian negotiators opened technical talks in Qatar, the first substantive contact since tensions flared over the Strait of Hormuz.
Sentiment sits at Extreme Fear as investors weigh geopolitical risk against a wave of new AI hardware and model releases.
Etched raised $800 million and showed a working inference chip, betting that transformer-specific silicon can undercut Nvidia's general-purpose GPUs.
DiffusionGemma applies diffusion techniques to text generation, a departure from the autoregressive approach dominating the field. Paired with Gemini 3.5 Flash, Google is signaling a two-front push on both novel architectures and low-cost, high-throughput inference.
Etched's $800 million round and functional inference chip mark the strongest test yet of Nvidia's dominance. Purpose-built transformer accelerators could reshape unit economics if they deliver the throughput gains their backers promise.
Washington restored global access to Fable 5 and Mythos 5, reversing the restrictions that followed the earlier jailbreak crisis. The move reflects growing confidence in Anthropic's guardrails and a broader loosening of AI export policy.
The Great Inference Land Grab
Read full analysis arrow_outwardGoogle launched DiffusionGemma, an experimental mixture-of-experts model using diffusion-style parallel token generation for local autocomplete, inline editing, and code infilling, running several times faster than existing Gemma...
The Commerce Department lifted export licensing requirements on Anthropic's Fable 5 and Mythos 5 models, restoring global public access starting July 1 after 19 days offline. Anthropic is also proposing an industry-wide framework for...
AI chip startup Etched raised $800M, unveiled a functioning inference chip, and secured more than $1B in signed customer contracts. The round is backed by Jane Street and TSMC-linked investors, signaling growing appetite for specialized...
OpenAI introduced GeneBench-Pro, a benchmarking tool designed to evaluate AI performance on genomic tasks. The release outlines capabilities and applications for biological research, deepening OpenAI's push into scientific evaluation.
Indirect technical negotiations between US and Iranian teams commenced in Qatar this week, the first diplomatic engagement since the February 28, 2026, outbreak of the Iran war. Iran's top negotiator said high-level talks will not...
The UN is calling on donors to fill a $100 million funding gap for UNRWA, the agency for Palestinian refugees, as of July 1, 2026. Secretary-General Antonio Guterres warned that millions of Palestinians are at risk due to the shortfall...
President Trump's sweeping changes to US student loans became effective on July 1, 2026, imposing higher monthly payments and tighter borrowing limits for many borrowers. CNN reports experts warn the overhaul will ripple across...
Rights groups filed a complaint on July 1, 2026, accusing Ghana of complicity in unlawful US deportations under Trump's third-country agreements. The complaint involves 27 of at least 60 immigrants deported to Ghana since September...
Two deadly heat waves have struck Europe in recent weeks, with dangerous temperatures now spreading across the US as of July 1, 2026. An extreme heat warning is in effect for parts of Indiana, Kentucky, and Ohio, with heat index values...
Bolt.new lets you create, run, edit, and deploy full-stack web applications directly from prompts, with real-time changes as you iterate. Use its prompt-run functionality to spin up working prototypes fast, ideal for startups testing an idea before committing engineering hours. It is free to use.
Try it arrow_outwardRocket transforms a single prompt into a production-ready application, generating clean code with backend and integrations included. It ships with GitHub and Netlify integration for one-click deploys, making it useful for non-technical founders and developers who want an MVP without stitching services together. It has a freemium tier with paid plans from $25/month.
Lab report scienceSuperhuman AI is a suite of AI features built to help you clear your inbox and hit inbox zero faster, drafting replies and triaging mail automatically. Use it to cut time spent triaging email so you can batch-process messages instead of reacting to each one.
Try it arrow_outwardSciSummary uses a custom-tuned model to digest dense scientific articles into quick, accurate summaries, and you can even submit papers by email. Students and busy researchers can use it to extract the core findings of a study before deciding whether to read the full text. It offers a free trial with paid plans from $6.99/month.
Lab report scienceCoda AI consolidates your tool stack into a single workspace, streamlining workflows and cutting the cost of running separate apps. Use it to unify docs, tables, and automations so a small team can operate without juggling multiple subscriptions.
Try it arrow_outwardEchoTik.live is a TikTok data tool that tracks target creators and videos, analyzes trending topics and products, and compares performance insightfully. Creators and brands can use it to spot hot products and refine posting strategy before competitors catch the trend. It has free options.
Lab report scienceTest DiffusionGemma locally. Wire Google's diffusion model into your editor for autocomplete and code infilling, then benchmark latency against your current autoregressive assistant.
Spin up a project in bolt.new. Scaffold a working app from a single prompt, then hand off the export to Rocket to iterate on the frontend without touching boilerplate.
Re-integrate Fable 5 or Mythos 5. With global access restored July 1, re-run your agent evals against Anthropic's models and confirm your prompts survived the 19-day gap.
Four launches in a single week signal how fast the model and silicon layers are compressing. Etched's working chip and $1B in contracts show the hardware moat is contestable, while DiffusionGemma hints the autoregressive orthodoxy is not settled. The reopening of Fable 5 and Mythos 5 restores a stack builders had already learned to trust. Move now: cheaper inference and restored access reward teams that ship before the next reshuffle.
AI analysis, world news, markets, and tools. One briefing, delivered free.