Google Gemini and DeepMind

Oct 9

On October 8, 2026 the DeepMind Institute published "Bending the Curve of Discovery", an essay built on 15 million Gemini interactions, 2,600 specialized models and a survey of 637 scientists. explainx.ai walks through the findings, the bottleneck shift from computing to the lab bench, and what is claimed versus independently verified.

Oct 9

On October 8, 2026, Google Research said its AMIE diagnostic chatbot study at Beth Israel Deaconess Medical Center was published in The Lancet. The headline "matches doctors in 90% of cases" hides a more careful result: here is what was measured, where physicians still won, and what to ignore.

Oct 8

Google launched AI Edge Foresight, an experimental macOS meeting companion that runs EmbeddingGemma 2 and Gemma 4 locally. It captures system audio and your mic, expands your shorthand notes, and searches your files without a cloud subscription. Here is what is confirmed and what is not.

Oct 8

At Gemini at Work 2026, Google Cloud introduced the Gemini agent, a single cloud-hosted agent that answers, works autonomously, and writes code, can run Claude models underneath, and is in private preview for enterprises. explainx.ai breaks down what shipped and what is still unproven.

Oct 7

On October 7, 2026, Biohub, the Department of Energy, the NIH and new funding partners announced an expansion of the Virtual Biology Initiative. Reuters reports Meta, Google DeepMind and Isomorphic Labs are investing $300M. Here is the money, the data plan, the embargo terms and the open questions.

Oct 7

Google Playground lets US adults build, remix, and share browser games from plain-text prompts. explainx.ai breaks down what Google confirmed, what it left out, and how it compares with Roblox, Unity, and prompt-to-game startups.

Oct 7

On October 7, 2026, Google opened its SynthID verification site to everyone, after a limited beta for journalists and media professionals. Here is what it can and cannot tell you, which file types it accepts, and how it fits with C2PA and text watermarks.

Oct 6

Google DeepMind released EmbeddingGemma 2, an Apache 2.0 embedding model built on Gemma 4 that puts text, code, images, video and audio into one 768-dimension space and runs on a phone. Here is what is in the box and how to use it.

Oct 6

Google and Constellation announced a 20-year agreement to fund uprates at Constellation nuclear units, adding 890 MW to the PJM grid by 2032, with a separate 2.7 GW supply deal. Here is what kind of deal it is and how it compares to the other hyperscaler nuclear announcements.

Oct 6

Google released Nano Banana 2.1 on October 6, 2026, an image generation and editing model at Flash-tier speed that costs about half as much as Nano Banana 2. The old model shuts down on October 29, so API users have a three-week migration window. Here is what changed and what to test first.

Oct 5

On October 1, 2026, Google stopped accepting product vulnerability submissions to its Open Source Software Vulnerability Rewards Program, saying automated submissions had surged and most were not valid. Here is what is paused, what still pays, and what it means for maintainers and researchers.

Oct 4

Google is pulling model access from its lowest Gemini tiers. Starting October 9, 2026, free users keep only Flash-Lite, and AI Plus subscribers lose Gemini Pro while keeping Flash. AI Pro and AI Ultra are unchanged. Here is the plan-by-plan breakdown and how to adapt.

Oct 3

A Google Research preprint describes Cogentic, a Gemini-based multi-agent harness that produced novel results on five open problems in online learning, auctions, and mechanism design. Domain experts verified the proofs. Here is what Cogentic is, which five problems moved, and what a builder should trust.

Oct 2

Oct 1

On September 30, 2026 Google DeepMind announced Gemini 4 Argon. Fairwind first, 1M output tokens, $2/$10 intro. This update adds what the Hacker News thread actually argued: harness, compaction, and still-no-API.

Oct 1

On September 30, 2026 Google said Gemini chat Skills will replace Gems. Here is who can use them today on a free personal account, what Workspace still has to wait for, what happens to your Gems, and why this is not the same as Claude or Cursor agent skills.

Oct 1

On September 30, 2026, Bloomberg reported internal skepticism that Gemini 4 scores well on industry coding benchmarks but does less well when Google employees put it to work. Google disputed that characterization. This post separates anonymous-source claims from Google's denial and from the coding split already published in Argon's own scorecard.

Sep 30

On September 30, 2026, Google DeepMind introduced SynthID Bio: watermarks that survive from a digital protein design into a physical molecule, without wrecking wet-lab function. This is provenance for biology, not a new drug model.

Sep 27

A September 27, 2026 wave of arena clips is being read as Gemini 4 Pro: a floatplane against Claude Opus 5.5, a motion comparison against GPT-6 Astra, and one-shot 3D scenes. Google has only confirmed early post-training. This is the demo catalog, with the claims that do not hold.

Sep 27

A September 26, 2026 TechCrunch report describes Google testing a Flipkart Buy button inside Gemini and AI Mode for Indian users — checkout that stays visibly Flipkart-branded in the AI UI rather than routing through Google''s Universal Commerce Protocol end-to-end. Categories are narrow for now, a broader rollout is expected around October 2026, and Amazon product listings can appear without an equivalent one-tap buy path. explainx.ai breaks down what that split means for builders, merchants, and the agentic-commerce stack India is assembling.

Sep 26

Pranav Reddy's September 26 floatplane clip and a September 28 leaked benchmark table both label a rumored Gemini 4 Pro checkpoint ahead of Claude Opus 5.5 and GPT-6 Astra — on WebGL demos and on agent scores like DeepSWE and Terminal-Bench 2.1. Google has not confirmed the SKU or the numbers. explainx.ai separates verified facts from arena nicknames and screenshot math.

Sep 25

Google announced three Chrome features aimed squarely at students on September 24, 2026: Gemini can now analyze podcasts and non-YouTube video, generate interactive quizzes from your open tabs and Google Docs, and sync a tab's exact scroll position across devices. Here's what shipped, what's still US/India desktop-only, and how it fits the rest of Google's AI study push.

Sep 25

At The Information AI Agenda Live Summit on September 24, 2026, new Google DeepMind leader Koray Kavukcuoglu said Gemini 4 entered post-training and that Google intends to ship an early post-training build as soon as possible — potentially well before end of 2026 — after Gemini 3.5 Pro never launched. Here is what post-training means, why Google skipped 3.5 Pro, and how builders should prepare API and agent routes.

Sep 25

Google confirmed the Transporter-18 launch: a Planet-built prototype carrying four TPUs is in orbit and responding. The chips still only burst about 15 minutes before they must cool. This is a hardware survival test, not Gemini from space.

Sep 25

On September 25, 2026, Google Research announced a unified multi-agent framework for temporally consistent long-form video. It bundles four papers into one pipeline that tracks world state, plans globally, generates segment by segment, and critiques its own output. Here is what each part does and what you can borrow today.

Sep 25

On September 24, 2026, Wiz announced Scan for Good with Google DeepMind — free authorized scanning of public-facing critical infrastructure, nonprofits, and public services using Wiz Red Agent and Gemini 3.8 Flash Cyber, human validation, and CISA collaboration. Early results included hundreds of high-severity findings across rail, hospitals, archives, and open-source repos.

Sep 24

Google DeepMind released two new text-to-speech models in the Gemini API and AI Studio: Flash for creative direction and character design, Flash-Lite for cost-efficient scale. They add prompt-based voice design, 2,000+ ready voices, 100+ languages, voice replication from 30 seconds, and a first-place claim on Hume AI's Voice Design Benchmark.

Sep 23

Deploying DiffusionGemma-Jev — the open-source Jev clone built on Google's diffusion Gemma model — used to mean provisioning your own GPU. A new project, djev-run, cuts that down to a single gcloud command that spins up a Jev API-compatible endpoint on Cloud Run, scaling to zero when idle. Here's what it actually does and what it costs to run.

Sep 23

Google Labs shipped "CC," an experimental agent with its own verified Google account that up to six household members can share, built to handle school permission slips, shared calendars, and meal planning. The launch thread spent as much time confused about the name — which reads, to developers, exactly like Claude Code's shorthand — as it did discussing the product itself.

Sep 21

AX is Google's new declarative control plane for running agentic workloads on Kubernetes, built on "Agent Substrate" and released under Apache 2.0 on github.com/google/ax. It topped Hacker News at 179 points and 74 comments, with commenters split between "finally, real infrastructure for agents" and "remember Gemini CLI."

Sep 21

Four months after teasing an "intelligence system" for laptops, Google put Googlebook up for preorder on September 21, 2026. explainx.ai breaks down the confirmed specs, what Google still hasn't said, and why this launch is really a shot at Microsoft's Copilot+ PC program and Apple's on-device AI strategy.

Sep 19

During a May 2026 cybersecurity evaluation run by Irregular, Gemini-based agents were meant to attack fictional target companies in an isolated test environment — but a configuration error gave them real internet access, and the fictional targets shared names with real businesses. Gemini guessed passwords into one system and used credentials found in a public repository to access two more, then stopped on its own once it realized the systems were real. Google didn't disclose until the Wall Street Journal asked, four months later.

Sep 18

Google updated its Gemini API managed agents with a new harness, antigravity-preview-09-2026, bringing the Antigravity coding agent's tools and behavior into AI Studio and the Interactions API on Gemini 3.8 Flash. Two new APIs ship alongside it: Files, for moving data in and out of an agent's sandbox, and Credentials, which lets an agent call GitHub or Slack without the model ever seeing the actual token.

Sep 18

Google Research introduced a generative UI system built specifically for classroom use on September 18, 2026 — teachers describe a topic and get a guided, interactive simulation back, constrained by learning-design guardrails rather than an open-ended UI generator. A 30+ item STEM sample library ships alongside it, and a developer publicly noted he'd built a similar tool months earlier with off-the-shelf LLMs and custom UI tooling.

Sep 17

Google gave three companies early access to its newest Gemini Flash models to test agentic video understanding on real workloads. The results are specific enough to design against: a 97% median token cut, a 0.967 F1 score, and a 65% accuracy improvement, each from a different architectural choice.

Sep 16

Dream-RSI, from Google, University of Maryland, and Google DeepMind, has an AI agent replay its own past discovery attempts in a cheap offline simulator, then redeploy the improved search strategy online. It cut discovery cost sharply across algorithm engineering, math optimization, and GPU kernel work — while leaving the underlying model's weights untouched.

Sep 16

Demis Hassabis and Shane Legg announced the DeepMind Institute on September 16, 2026 — a platform for interdisciplinary research on what AGI means for the economy, science, and society. It launched with five essays, including one on eleven economic policies for AI-driven disruption. Here's what's actually in it, and what the reaction on X pushed back on.

Sep 16

Gemini 3.8 Live and 3.8 Live Extended Thinking are Google's newest voice models, built for near real-time dialogue with background tool calls and live progress narration. One leads a speech-quality index outright; the other trades some of that fluency for deeper multi-step reasoning.

Sep 11

Google DeepMind, Primordial Soup, and Story Syndicate's Telluride-selected short Love, Rendered recreates Burt and Ethelle Shatz's unfilmed first meeting using restored archival photos plus pose and performance control. explainx.ai breaks down the workflow, the human-in-the-loop design, and what builders should take from a reconstructed memory that still needs provenance.

Sep 11

Google announced its largest single investment in Europe on September 9-10, 2026 — a €13 billion ($15 billion) package for Finnish AI infrastructure, anchored by a 22-year contract to buy up to 50% of the output from Finland's Loviisa nuclear power plant.

Sep 11

Sep 11

Sep 10

Sep 10

Sep 7

Sep 5

Sep 5

Sep 3

Sep 3

Sep 2

Sep 2

Sep 2

Sep 1

Sep 1

Aug 31

Aug 29

Aug 28

Aug 28

Aug 27

Aug 27

Aug 27

Aug 27

Aug 24

Aug 22

Aug 21

Aug 20

Aug 18

Aug 17

Aug 14

Aug 13

Aug 13

Aug 12

Aug 7

Aug 6

Aug 5

Jul 30

Jul 30

Jul 21

Jul 16

Jul 14

Jul 13

Jul 11

Jul 7

Jul 3

Jul 3

Jul 1

Jun 30

Jun 30

Jun 29

Jun 27

Jun 27

Jun 24

Jun 20

Jun 19

Jun 19

Jun 17

Jun 15

Jun 14

Jun 13

Jun 11

Jun 11

Jun 8

Jun 4

May 30

May 26

May 22

May 22

May 20

May 20

May 13

May 13

May 12

May 8

Apr 30

Apr 29

Apr 24

Apr 15