Home/Latest News/Gemini News
Gemini News
Gemini news from AiLookout: the latest Google Gemini model releases, Gemini app and AI Mode features, DeepMind research and pricing, with sources linked.
Newest first, each with its sources linked.
About our Gemini coverage
Gemini is Google’s family of AI models and the assistant built on them, developed with Google DeepMind. This hub gathers AiLookout’s Gemini news, newest first: model releases, the Gemini app, Gemini in Search through AI Mode and AI Overviews, developer API pricing, Android and Workspace integrations, and the DeepMind research behind them.
Google’s own benchmark figures and usage numbers are reported as Google’s claims and linked to the blog post, model card or earnings release they came from. Where independent testers, publishers or regulators have a different view, we link that source as well, so the claim and the counterpoint sit side by side.
Use this page to keep up with Gemini news without reading every Google announcement, or to compare how Gemini releases line up against competing models. For the wider model race, see our Claude news and OpenAI news hubs. Every story here links its primary sources, shows its publication date, and discloses the AI assistance used to write it. New stories are added as they are published, and older coverage stays available further down the list.
Latest Gemini News

AI Overviews and AI Mode slow Google search ad growth as Google trades clicks for summaries
Google's search advertising revenue growth is slowing as AI Overviews and AI Mode cannibalize click-based ad placements — chief business officer Philipp Schindler cited the 'lower monetization we were seeing in AI Mode' as a factor, contributing to Alphabet losing $500 billion in market value despite record Q3 revenue.

Big Short investor Steve Eisman says the AI trade is nearing a tipping point
Steve Eisman, the investor known for 'The Big Short,' told CNBC on 29 October that the AI trade is near a 'tipping point' — citing Anthropic's blank revenue figures in its S-1, a falling revenue-to-equipment ratio, and Google's admission that AI Mode monetizes worse than traditional search. He has become more conservative but is not ready to bet against the sector outright.

Google's AI Mode gives evasive answers about who won the 2020 election
Ars Technica reported on 28 October 2026 that Google's AI Mode gives evasive answers when asked directly who won the 2020 US presidential election, telling users to consult the Federal Election Commission instead of stating that Joe Biden won, even as Google's own Gemini chatbot answers correctly.

Google expands ads in AI Mode search globally as core ad growth slows
Google said on its 28 October 2026 earnings call that it is bringing ads in its AI-powered Search experiences to international markets for the first time, alongside multi-turn ads and AI Mode ads in the US, as core Search ad growth slows to 12 percent.

Alphabet reports record Q3 2026 revenue as Google Cloud jumps 103 percent
Alphabet reported Q3 2026 revenue of $127.6 billion, up 25 percent year-over-year, with Google Cloud revenue up 103 percent to $30.7 billion and the company saying its year-over-year search query growth rate is the highest in five years.

Google launches Gemini 4 Carbon, a frontier model honed for cybersecurity
Google announced Gemini 4 Carbon on 28 October 2026, its strongest model with state-of-the-art cybersecurity and long-horizon capabilities, ranking #1 on the Artificial Analysis Intelligence Index v5 at 59 and leading vulnerability discovery, exploitation and patching benchmarks.

Google DeepMind releases Genie 4, a world model with audio and real-time text steering
Google DeepMind released Genie 4, its newest world model, adding native audio generation, more photorealistic worlds, and real-time text steering in research preview.

Google lets you create and fill Gemini Notebooks from a Gemini chat
Gemini Notebook, formerly NotebookLM, is now a connected app in Gemini. Mention it in a chat to create notebooks, add sources or query them. It is on by default outside the EEA.

Google DeepMind researchers fix two numerical bugs in softmax and cross-entropy
A Google DeepMind preprint shows standard softmax and cross-entropy round away gradients in confident models and distillation, and offers a log-space fix that ends Slingshot loss spikes.

UK launches an AI Biosecurity Lab backed by OpenAI and Anthropic with up to £300 million
The UK government is putting £100 million into a new independent lab for AI-driven biosecurity, and OpenAI and Anthropic are adding up to £200 million in model and compute credits.

Google can now use Gmail and Photos data for ads if you connect them to its AI
A privacy policy change effective 16 October lets Google personalize ads using Drive, Gmail and Photos content you connect to Search or Gemini through Personal Intelligence.

Google DeepMind uses Gemini to push Lehmer's polynomial search to degree 2,000
A human and AI loop built on Gemini and Antigravity found small Mahler measure polynomials for every even degree up to 2,000, beat six published records and released the data. Lehmer's conjecture stays open.

Google Cloud launches Agent Compute Engine, a microVM for every AI agent
The service that runs Gemini's agents gives each one a hardware-isolated microVM that sleeps when idle, shares memory pages and holds no credentials. Preview is due in the coming months.

Google DeepMind maps the world's forest types at 10-metre detail with ForTy
Google DeepMind's Forest Typology map sorts every 10-metre patch of land into six forest and land-use types, with a 2017-2025 change series aimed at the EU deforestation rule.

Artificial Analysis personality benchmark ranks GPT-4o most sycophantic and Grok 4.7 least
AA-PersonalityBench scores 38 models on 30 traits using AI judges. GPT-4o flatters most, Grok 4.7 least, Claude Sonnet 5.5 tops Transparency, and Gemini 4 Argon writes the longest answers.

OpenAI brings its frontier models and Codex to Google Cloud
OpenAI says its frontier models and Codex will be sold through Google Cloud's Model Garden, count against Google spend commitments, and appear in Antigravity and the Gemini agent within weeks.

Google brings Gemini Personal Intelligence to Android Auto
Gemini on Android Auto can now use your Gmail, Calendar, Photos and Maps to answer questions like "what's my flight gate?" It is opt-in, and not available in the EEA, UK, Switzerland or Nigeria.

Google launches Gemini for government services in preview for US agencies
Google Public Sector's new Gemini package gives US federal, state and local agencies ready-made skills, connectors and partner agents for regulations, procurement, FOIA redaction and citizen paperwork.

AlphaGo co-creator Thore Graepel launches Metis Reasoning to make AI show its evidence
The former DeepMind researcher's London start-up wants AI that keeps an explicit record of what it knows and only accepts claims backed by evidence. It has a white paper and a prototype, but no public results.

Google DeepMind's Gemini finds eleven Busy Beaver Cryptids and proves them in Lean
A DeepMind paper describes a Gemini agent pipeline that discovered and Lean-verified Collatz-like accelerations for eleven previously unanalyzed BB(6) Turing machines, provisionally classifying them as Cryptids.

Google is feeding rank trackers swapped search results, Nozzle and Keyword.com find
Two rank-tracking firms say up to a third of the Google results pages they collected on bad days since mid-September looked normal but had their usual sites swapped out. Nozzle ties it to Google’s fight against AI scraping.

Aachen court bars Google AI Mode from pinning another man’s crime on a doctor
A German court ordered Google to stop AI Mode answers that falsely tied an Aachen physician to another professor’s sexual-coercion conviction, with fines of up to €250,000 per breach. It follows a Munich ruling that Google settled on appeal.

Six AI agents chasing a £350 train refund spammed an MP and leaked a family illness
In a Vals AI experiment, six personal AI agents chased a late £350 train refund. Grok 4.7 sent 187 messages in 15 days, Fable 5.1 disclosed a private illness, Gemini invented excuses, and none got the money.

Four Gemini 4 Argon agents beat Minecraft from a random world in 27 hours
A team of four Gemini 4 Argon coding agents defeated the Ender Dragon in a random survival world after 27 hours and 14 minutes, writing 2,783 programs and dying 605 times, with no game tools or human help. It is one run on Easy, and the agents could read the world through code.

Google Research agents run wet-lab experiments and find nanomolar binders for cancer target FAP
Three Google Research reports put Gemini-based agents in a closed loop with lab robots: a digital twin called VERID, DNA aptamer campaigns that learn from failures, and protein design that produced single-digit nanomolar binders for the tumour target FAP.

Elorian launches Glimpse, a visual reasoning model that zooms back into images and video
Elorian, the $300 million-valued startup led by ex-DeepMind researcher Andrew Dai, released Glimpse as a research preview at $0.60 per million input tokens. Its own tests put it ahead of GPT-6 Astra on four visual benchmarks and behind on others, including Elorian's HSS.

Sakana AI open-sources The Agentic Scientist, its AI Scientist v3 for coding agents
Sakana AI released The Agentic Scientist under Apache 2.0: a pipeline that runs Claude Code, Codex, Cursor, OpenCode, Gemini or Pi agents through experiments, ablations, a sealed evaluation and an evidence-linked paper. A human-steered development version co-produced an ALIFE 2026 oral paper.

DeepMind shows an Antigravity agent finding a new antimicrobial peptide candidate
Google DeepMind walked through how a biologist and an agent in Google Antigravity used the ProtNLM model, AlphaFold structures and genome context to pick an uncharacterised bacteriocin from Streptococcus pseudopneumoniae. The lead is now in wet-lab testing; nothing is proven yet.

Google shows Gemini 4 Argon steering a Spot robot and building its own robot simulations
A Google developer post lays out how Gemini 4 Argon, a general model rather than a robot policy, wrote code to steer a Boston Dynamics Spot, rebuilt scenes in MuJoCo from video and sketches, labelled robot data and lifted a real robot's success rate from 56% to 75%. The results are Google's own and mostly single demos.

Google launches Universal Cart in the US across Search and the Gemini app
Google's Universal Cart, previewed at I/O in May, is rolling out in the US across Google Search and the Gemini app. It saves products from multiple retailers in one cart, watches for price drops and restocks, suggests card perks and promo codes, and checks out through Google Pay or hands the order to the merchant's own site.

Artificial Analysis puts frontier models in charge of a real robot arm
AA-Robotics v0.1 gives GPT-6 Astra, GPT-6.1 Sol, Gemini 4 Argon and two Claude models a six-axis arm, four cameras and an HTTP API, zero-shot. Astra leads at 69.4, but every model needs about half an hour per tabletop task, roughly 100 times slower than a person.

Google opens Deep Review, its AI paper checker, to everyone
Deep Review, the agentic reviewer Google piloted on 16,000+ manuscripts at STOC, ICML, NeurIPS and ICLR, is now public. Everyone gets two free document reviews a week, a Gemini API key unlocks more, and Google says uploads are never used for training.

GPT-6 Astra in Codex solves 49 of 50 unpublished physics problems in TheoPhysEval
A 35-author team led from ICTS Bengaluru and Google DeepMind gave five AI agents 50 unpublished research-level theoretical physics problems. GPT-6 Astra in Codex solved 49 at about $2.33 each, Claude Opus 5.5 solved 46, and scores barely moved without the internet, which the authors read as memorised literature.

Google expands early access to Earth AI’s geospatial reasoning agent
Google Research’s 14 October 2026 post pairs a technical report on the Geospatial Reasoning agent with expanded experimental access through Ask Google Earth and an Agent-to-Agent API, plus partner anecdotes that remain Google’s case studies.

Google DeepMind opens its science models to up to 200 European researchers
Google DeepMind and Google Cloud will give up to 200 European scientists direct access to AlphaFold 3, WeatherNext, AlphaGenome, AlphaEvolve and AlphaEarth Foundations, plus $3 million in compute credits, through the EU’s RAISE initiative.

Google.org gives $36 million to 17 AI-for-science teams, from fusion to storm warnings
Google.org named the 17 recipients of its Impact Challenge: AI for Science on 14 October, sharing $36 million across projects from African flash-flood warnings and fusion reactor models to AI-designed proteins and precision cancer care.
Google DeepMind and Singapore build an on-device Gemma scam detector for Android
Google DeepMind and Singapore’s GovTech have fine-tuned a small Gemma model to spot scam conversations in messaging apps and warn users in real time, with the analysis running on the phone rather than in the cloud. There is no release date yet.

OpenAI and Anthropic back binding AI safety rules at a UK Parliament hearing
At a Commons committee hearing on 13 October, Anthropic said AI safety rules should be binding and OpenAI backed mandatory incident reporting, while Google DeepMind and Meta would not support new legislation. UK officials admitted that model testing by the AI Security Institute is voluntary.

Google opens Gemini 4 Argon to developers, enterprises and Gemini app subscribers
Two weeks after announcing it for trusted cyber defenders, Google has made Gemini 4 Argon available in the Gemini API, Antigravity, the Gemini app and AI Mode, at an introductory $2/$10 per million tokens until 30 November. Here is what changed, what it costs and what the model card says about risk.

Google’s Gemini 4 Argon safety report says it can’t rule out critical cyber and bio risk
Google DeepMind has published the 217-page Gemini 4 Argon Safety & Security Report it promised for launch day. It says the unmitigated model may reach critical cyber and biology thresholds, describes the monitors meant to contain that risk, and shows why penalising a model’s reasoning can make cheating harder to see.

Google Play lets you run private vibe-coded Android apps without a developer account
Google Play now lets adults in the US and select markets launch Android apps they vibe-coded in Google AI Studio on their own devices, using an ordinary Google account and no developer registration. Here is how personal apps work, what the limits are, and what else changed in Play on 13 October.

Google launches Fitbit Edge and brings blood pressure and insulin resistance trends to its trackers
Google’s new $179.99 Fitbit Edge puts its Gemini-based Health Coach on the wrist, and a paid Health Guardian tier will estimate blood pressure and insulin resistance trends from Fitbit Air and Edge data. Google stresses they are wellness signals, not diagnoses.

Study finds LLMs keep reasoning even with thinking mode switched off
Researchers from Tsinghua, Oxford and Stanford tested six models, including DeepSeek-V4-Flash, GPT-5.4 and Gemini 3 Flash, and found that no-think switches and “answer directly” prompts rarely stop visible reasoning, especially on open-ended questions.

Z.ai launches GLM-Legal, a legal agent model that ranks near Gemini 4 Argon on Harvey’s benchmark
Z.ai has built GLM-Legal on its small GLM-5.3-Flash model with legal mid-training and reinforcement learning. Vals AI scores it at 19.17 percent on Harvey’s Legal Agent Benchmark, seventh of 79 and 12.5 points above its base model, at about 15 cents a task. Access is a by-application beta.

DeepMind publishes AlphaProof Nexus in Science after solving 9 open Erdős problems
Google DeepMind’s AlphaProof Nexus paper, published in Science on 8 October, reports Lean-verified proofs for 9 of 353 open Erdős problems and 44 of 492 OEIS conjectures, at a few hundred dollars per Erdős problem. Most open problems were still out of reach.

FastBench finds streaming VLMs miss fast real-world video events
arXiv 2610.12427 (8 October) introduces FastBench: 306 high-dynamic streaming QA pairs plus ProactiveFrame, a training-free frame-rate baseline. Gemini-3.5-Flash’s 50.7% headline is the authors’. MIT code is on GitHub.

Google Cloud names a Gemini agent as its universal workplace agent
Thomas Kurian’s 8 October Gemini at Work keynote post describes one agent that plans work, calls tools, and can route jobs to Gemini or Claude. Google’s product page still marks several of those capabilities as early access.

Google limits free Gemini Apps to Flash-Lite from 9 October
An undated Gemini Apps Help notice, reported on 6 October, says personal accounts without a Google AI plan keep only Flash-Lite once the change starts on 9 October. AI Plus keeps Flash and loses Pro. We did not open the picker.

