Skip to content

Google / Gemini

AI at Google and DeepMind: the Gemini family, Veo video models, research and the product ecosystem.

Latest picks

401–420 of 524

May 20Wednesday

TechCrunch · AI

Google introduces Gemini Spark, a 24/7 agentic assistant with Gmail integration, at I/O 2026

Google introduced Gemini Spark at I/O 2026 as a 24/7 agentic personal assistant with Gmail integration; the RSS snippet says it uses Gemini base models and an agentic harness from Google Antigravity, but the post does not disclose pricing, rollout timing, or supported Gmail actions.

Why it matters: Gemini Spark is a 24/7 personal assistant built on Gemini models and the Antigravity agent framework, with Gmail access to read mail and handle tasks.

AI HOT (Curated Pool)

I/O 2026: Welcome to the autonomous Gemini era

Google announced at I/O 2026 that Gemini is moving into an autonomous agent phase, with the post saying it can manage email, schedule calendar items, and generate reports automatically, but it does not disclose model parameters, launch timing, or pricing.

Why it matters: Google said at I/O that Gemini moves into autonomous agents that handle email, calendars and reports, but gave no parameters, launch date or pricing, so it reads as a direction, not a product.

AI HOT (Curated Pool)

Google AI subscription updates from I/O 2026

Google announced a $100 AI Ultra subscription at I/O 2026 and added new features and benefits for existing Google AI Plus, Pro, and Ultra subscribers.

Why it matters: Google added a $100-a-month AI Ultra tier at I/O 2026 and refreshed Plus and Pro, giving its subscription lineup three parallel price levels.

Hacker News front page

Gemini 3.5 Flash

The title names Gemini 3.5 Flash, while the RSS body only includes a documentation link; the Hacker News item has 196 points and 179 comments, and the post does not disclose parameters, pricing, or context-window details.

Why it matters: Google's Gemini 3.5 series shifts the model from answering questions to carrying out actions, drawing 196 upvotes and 179 comments on Hacker News.

AI HOT (Curated Pool)

Google AI Ultra plan gets a price cut and a new tier

Google cut the top AI Ultra plan from $250 to $200 per month and added a $100 monthly tier with 5x the Gemini app usage limit of Pro, 20TB of storage, early access to new features, and YouTube Premium under stated terms.

Why it matters: Google drops AI Ultra from $250 to $200 a month, adds a $100 tier, and raises usage limits to five times the Pro plan.

AI HOT (Curated Pool)

Gemini Spark: 24/7 Autonomous AI Assistant

Gemini Spark runs personal-agent tasks autonomously in the background, including when a user’s phone and laptop are off; the post says it asks for user approval before major actions, but it does not disclose launch timing, pricing, or task scope.

Why it matters: Google previews an always-on personal AI agent that keeps running tasks after your phone and laptop are off, asking before big actions.

AI HOT (Curated Pool)

Google releases Gemini 3.5 Flash with output speed about 4x GPT-5.5

Google introduced Gemini 3.5 Flash at I/O 2026, with output speed reaching 289 tokens per second, about 4x faster than Claude Opus 4.7 and GPT-5.5 xhigh under the cited comparison.

Why it matters: Google's Gemini 3.5 Flash hits 289 output tokens per second, about four times faster than GPT-5.5 xhigh and Claude Opus 4.7.

AI HOT (Curated Pool)

Google launches Antigravity 2.0 platform, builds an OS in 12 hours

Google announced Antigravity 2.0 at I/O and demonstrated an agent building a runnable operating system from scratch in 12 hours, using 93 parallel sub-agents, more than 15,000 model calls, and 2.6 billion tokens, with API costs under $1,000.

Why it matters: Google's I/O demo used 93 parallel sub-agents, 15,000 model calls and 2.6 billion tokens to build a working OS in 12 hours for under $1,000 in API cost.

AI HOT (Curated Pool)

Google releases Gemini Omni for any-input-to-any-output generation and conversational video editing

Google released Gemini Omni and Omni Flash at I/O 2026, supporting text, image, audio, and video inputs and outputs, with conversational video editing; Omni Flash is available in Gemini App, Google Flow, and YouTube Shorts, while the post does not disclose the API launch date.

Why it matters: Google's Gemini Omni converts across text, images, audio and video, with a demo editing a video's character or background from one sentence.

AI HOT (Curated Pool)

Gemini 3.5 Flash launches as an efficient option for task handling

Google released Gemini 3.5 Flash and calls it its best model so far for fast, efficient task completion. The post does not disclose pricing, context window size, benchmark scores, or API availability conditions.

Why it matters: Google's Gemini 3.5 Flash targets fast, efficient everyday tasks and multi-step creative work, with pricing and context window still undisclosed.

AI HOT (Curated Pool)

Gemini 3.5 launches with frontier intelligence and real-world action

Google DeepMind introduced Gemini 3.5 with 3.5 Flash as the first release; the post says it is the company’s strongest model for agents and coding, but does not disclose parameters, pricing, or context window.

Why it matters: Google DeepMind's Gemini 3.5 series starts with 3.5 Flash, pitched at business workflows and coding, with agents and programming as the product focus.

AI HOT (Curated Pool)

Gemini Omni Released: New Progress in Multimodal Generation

Google DeepMind released Gemini Omni, starting with video generation; the post says it combines Gemini with its generative media systems, but does not disclose parameters, pricing, or availability.

Why it matters: Google DeepMind's Gemini Omni merges Gemini's understanding with its generative media stack, starting with video, aiming to turn any input into any output.

AI HOT (Curated Pool)

Google processes over 3,200 trillion tokens per month, up 7x year over year

Google said at I/O 2026 that it processed over 3,200 trillion tokens per month in May, while Gemini App exceeded 900 million monthly active users and the Nano Banana model generated more than 50 billion images cumulatively.

Why it matters: Google said at I/O that its platforms process over 3,200 trillion tokens a month, up 7x year over year, with the Gemini app past 900 million monthly users, a scale reference for its AI serving.

May 19Tuesday

Latent Space

[AINews] How to Land a Job at a Frontier Lab (on Pretraining)

Latent Space says Vlad Feinberg’s pretraining job-prep notes reduce frontier-lab readiness to kernel-level performance work: derive Chinchilla laws, compare dense and MoE architectures, code the solution in JAX, then write a Pallas kernel that beats jax.lax.ragged_dot for F > D by fusing up/down projections.

Why it matters: A job-hunting note from Vlad Feinberg lays out what pretraining work at a frontier lab actually tests: low-level performance tuning, deriving Chinchilla scaling laws, and comparing dense models with MoE architectures.

Synced · WeChat

Recent LLM Architecture Changes: From Gemma 4 to DeepSeek V4

Jiqizhixin translated Sebastian Raschka’s blog on recent LLM architecture changes, covering long-context cost reductions in Gemma 4, Laguna XS.2, and ZAYA1-8B; the article states that Gemma 4 E2B saves about 2.7GB of KV cache at 128K context with bfloat16 precision.

Why it matters: Raschka walks through architecture changes in Gemma 4 through DeepSeek V4, all aimed at cutting long-context memory cost.

Financial Times · Technology

Google makes chip push with Blackstone-backed AI cloud group

A Blackstone-backed AI cloud group is set to receive a $5 billion investment to bring 500MW of data center capacity online next year; the post does not disclose the Google chip terms or deployment structure.

Why it matters: A Blackstone-backed AI cloud group raised $5bn and plans 500MW of data center capacity next year, with the Google chip tie-up still undisclosed.

AI HOT (Curated Pool)

Google and Blackstone form AI cloud company with $5B initial equity and 500 MW target by 2027

Google and Blackstone formed an AI cloud services company with Blackstone committing $5 billion in initial equity capital, an expected total investment of about $25 billion after leverage, and a plan to bring 500 megawatts of data center capacity online in 2027.

Why it matters: Google and Blackstone are forming an AI cloud company, with Blackstone putting in $5 billion of equity first against a $25 billion total, aiming for 500 MW of data centers by 2027.

Bloomberg Technology

Google, Blackstone to Create AI Cloud Firm With In-House Chips

Google agreed to create an AI cloud business with Blackstone using in-house chips to compete with CoreWeave; the post does not disclose ownership structure, investment size, launch timing, or chip specifications.

Why it matters: Google and Blackstone plan an AI cloud joint venture running on in-house TPUs against GPU cloud providers like CoreWeave, showing one route for big tech's own silicon into cloud compute.

May 18Monday

Synced · WeChat

ICML 2026: Huawei GTS proposes EDCO for dynamic curriculum fine-tuning

Huawei GTS proposed EDCO, a dynamic curriculum method that selects fine-tuning samples by inference entropy; prefix entropy estimation cuts per-sample scoring time from 2.24 seconds to 0.37 seconds.

Why it matters: Huawei GTS proposed EDCO, a dynamic curriculum fine-tuning method that uses reasoning entropy to judge sample difficulty and pick training data, with prefix-entropy estimation cutting per-sample scoring from 2.24 seconds to 0.37.

Google DeepMind

Google DeepMind adds Street View grounding to Project Genie

Google DeepMind has added Street View real-scene grounding to its experimental prototype Project Genie. Users can pick a US location, then pair it with a style and characters to generate a world.

Why it matters: With Street View imagery wired in, agents and robots can train and navigate in virtual environments that track real places.