AI NewsGoogle

Google AI Pro plans now bundle Colab with 80GB A100 GPUs, and Ultra adds H100s. Google’s Gemma team says that is enough to run Gemma 4 31B and tune smaller versions in a browser.
Teaching an open AI model new skills used to mean renting a pricey GPU (AI chip). Google’s Gemma team says Google AI Pro now includes Colab with 80GB A100s, enough to run Gemma 4 31B and tune smaller versions on your own data, in a browser.
It works like a rented workshop now bundled into your subscription. Ultra subscribers get even bigger H100 chips, so hobbyists and students can customize models without buying hardware.
Key Capabilities:

• Run it: Gemma 4 31B fits at full precision in 80GB of memory.
• Tune it: Full fine-tuning up to Gemma 4 E4B, and LoRA (lightweight tuning) up to 26B A4B.
• Bigger on Ultra: Ultra plans unlock H100 chips for heavier training runs.

Instead of switching between separate tools for email, docs, and code, Google’s new Gemini agent takes one plain-language goal and works across a company’s existing apps to deliver finished output.
Google Cloud unveiled the Gemini agent on October 8 at its Gemini at Work event, held at NASA’s Hangar One. It runs persistently in the cloud, so it keeps the same memory and context whether accessed from Gmail, Slack, or Microsoft 365, and it can spin up specialized “coworker agents” with their own identities for recurring roles. It also picks the best underlying model for each task, reportedly including Google’s own Gemini 4 Argon alongside Anthropic’s Claude Opus 5.5 and Sonnet 5.5.
Google said nearly 500 Cloud customers now process over 1 trillion tokens each per year, and about 90% of Fortune 100 companies use Gemini Enterprise. No separate pricing or full rollout date has been announced yet.
Key Capabilities:

• Persistent memory: Runs in the cloud, keeping one shared context across every app and device.
• Sub-agents: Can spawn specialized coworker agents with defined roles for recurring tasks.
• Model flexibility: Reportedly supports pulling in outside models, including Anthropic’s Claude, alongside Google’s own.

Playground turns a text prompt into a browser game, no coding required, while a deeper Unity-powered version called Spark is coming later this year for more complex projects.
Google and Unity launched Playground on October 7, available now to US users 18 and older at playground.google.com. Creators choose genre, perspective, and single- or multiplayer mode, then refine the game by chatting, with uploaded images converted into matching game assets. Generation runs on weekly token limits; Google One AI subscribers get a larger allowance than free accounts.
Games can be played and shared via link but can’t be sold or exported to desktop Unity at launch. Unity’s Spark, a closed beta coming later this year, is positioned as a tool for ongoing creation rather than one-shot game generation.
Key Capabilities:

• No coding needed: Describe a game in plain text; the AI builds and lets you refine it by chatting.
• Age-gated launch: Limited to US users 18 and older at launch.
• Not sellable yet: Games can be shared via link but not sold or exported to desktop Unity.

SynthID Detector, previously limited to journalists and researchers, is now open to anyone at synthid.com, checking files for an invisible watermark Google has applied to over 180 billion images and videos since 2023.
Google launched the public version on October 7, supporting uploads in common formats like JPG, MP4, and MP3, with a daily limit of roughly 10 checks per account. The detector also catches content from OpenAI, Nvidia, and Kakao, which all use the SynthID watermarking standard, with Apple support planned.
A clean result only means no SynthID watermark was found; it can’t confirm a file is real, catch AI content from models that don’t use SynthID, or detect open-weight models that skip watermarking entirely.
Key Capabilities:

• Multi-company coverage: Detects watermarks from Google, OpenAI, Nvidia, and Kakao, not just Google’s own tools.
• Free and global: Live now in English worldwide, with a daily per-account check limit.
• Known limits: Can’t flag content from AI tools that don’t use SynthID, including most open-weight models.

Nano Banana 2.1 costs half as much per image as its predecessor and jumped to #5 on an independent image-generation leaderboard, with notable gains in rendering legible text inside images.
Google released Nano Banana 2.1 on October 6, live now in the Gemini app, AI Mode in Search, Google AI Studio, and the API. A standard 1K image costs $0.0336, down from $0.067 for the prior version, and it supports resolutions up to 4K with up to 14 reference images for keeping a subject consistent across edits.
Google says it beats its own earlier models "across the board" on visual design, mask-based editing, and subject consistency, though those performance claims come from Google's internal testing rather than a fully independent benchmark.
Key Capabilities:

• Price cut: Roughly half the cost of Nano Banana 2 at every resolution tier.
• More references: Supports up to 14 reference images for consistent characters across edits.
• Leaderboard rank: Entered Text-to-Image Arena's leaderboard at #5.

EmbeddingGemma 2 can match a text search against an image, video, or audio clip, all using the same underlying system, and it's small enough to run entirely on a phone.
Google released EmbeddingGemma 2 on October 6, a 740-million-parameter embedding model, a type of model that converts content into a list of numbers so software can measure how related two pieces of content are. It handles text, code, images, video, and audio in one shared space, and runs in as little as 191MB of memory for text-only use on a Pixel 11 Pro.
It's released under the Apache 2.0 license, meaning it's free to use commercially, and supports over 100 languages. Developers can load only the parts they need, from a lightweight 270-million-parameter text-only version up to the full 740-million multimodal model.
Key Capabilities:

• Runs locally: Works entirely on-device, so search data never has to leave a phone or laptop.
• Modular size: Scales from 270M (text only) to 740M (all five data types).
• Free to use: Apache 2.0 license allows commercial use without a licensing fee.

Guided Vision in Gemini Live now describes what an Android camera sees in real time, built specifically for blind and low-vision users with help from accessibility group Aira.
Google launched Guided Vision on Android on October 1. Point a phone's camera at something during a Gemini Live session, and it describes surroundings, reads fine print, locates nearby objects, and gives spoken cues to help center what you're looking for. Users can ask detailed follow-up questions.
Google built it with input from over 1,000 testers in Aira's network and communities in India, Brazil, Singapore, and elsewhere. It works on Android 9 and up, but Google is clear it's not a replacement for a white cane or mobility aid.
Key Capabilities:

• Real-time description: Narrates surroundings and reads text aloud as the camera moves.
• Three ways to launch: Via Gemini settings, Android's Accessibility menu, or TalkBack.
• Not a mobility aid: Google explicitly says it isn't meant for navigation or obstacle detection.

Starting October 9, free Gemini users drop to the entry-level Flash-Lite model only, and AI Plus subscribers lose access to Pro entirely, pushing more people toward paid tiers.
Google confirmed the changes in an updated support document. Free users lose both Flash and Pro, keeping only Flash-Lite. AI Plus ($4.99/month) keeps Flash-Lite and Flash but loses Pro. AI Pro and Ultra subscribers see no cuts, and AI Pro also gains the Deep Think reasoning feature previously exclusive to Ultra.
The change follows Google's Gemini 4 Argon announcement and appears aimed at managing compute demand as newer models roll out. Plus subscribers will get an email with their specific cutoff date, which may differ from October 9.
Key Capabilities:

• Free tier shrinks: Flash-Lite only, down from Flash-Lite, Flash, and Pro.
• Plus loses Pro: Keeps Flash-Lite and Flash, but the Pro model moves behind a higher paywall.
• Pro gains Deep Think: AI Pro subscribers get the extended-reasoning feature once reserved for Ultra.

Instead of opening a separate Gem to reuse an instruction set, users can now type "/" plus a skill name directly in any Gemini chat.
Google launched Skills in the Gemini app on September 30, free for all users globally. Skills are reusable instructions, for tasks like prepping a presentation or matching a writing style, that can be stacked together in one conversation and triggered from the main prompt box.
Skills replace the older Gems feature, which Google will phase out starting November 17 for personal accounts, migrating existing Gems automatically. Business and education Workspace accounts keep Gems until March and June 2027, respectively.
Key Capabilities:

• Quick access: Typed with a "/" directly in chat, no separate Gem manager needed.
• Stackable: Multiple skills can be combined in a single conversation.
• Automatic migration: Existing Gems convert to Skills with no action needed from users.

Google's first Gemini 4 model claims the top spot on 13 of 19 published benchmarks, but it's going to cybersecurity testers first, not the public.
Google DeepMind announced Gemini 4 Argon on September 30. It targets long, complex work like coding and enterprise research, with an industry-leading 1 million token output limit. Google's own chart shows it beating GPT-6 Astra and Claude Opus 5.5 on most listed tests, including an 88.8% on LABBench 2.
Argon rolls out first to vetted cyber defenders through the Fairwind Program, without cyber guardrails, while a public version with safety restrictions comes later for paid API and AI Ultra users. Google says internal teams already used it to free 300 TiB of data center memory.
Key Capabilities:

• Long output: A 1 million token output limit, four times more than most rival flagships.
• Cyber strength: Ties for first place on vulnerability remediation, scoring 68% on CWE-bench v1.
• Intro pricing: $2 per million input tokens and $10 per million output, with 95% off cached input.

A new slash command makes this AI coder research first and wait for a green light before editing files. It mirrors the planning mode already in Antigravity's command-line tool.
On September 26, Google announced /plan for Antigravity 2.0. Type it and the agent researches your codebase, then writes a step-by-step plan for review. It normally waits for your approval before touching any code.
Developers reacted with mixed feelings. A May forum thread asked where the IDE's plan mode had gone, and some mocked the return as a step backward. Still, a plan you can approve first means fewer surprise edits to your files.
Key Capabilities:

• Light version: Asking for a plan in a normal prompt gives a shorter, lighter plan.
• Your call: Unless your review policy is set to Always Proceed, you approve before execution starts.
• Edit the plan: You can comment on steps or ask the agent to shrink the scope

Google made Gemini 3.8 Live with Live Avatar generally available in Gemini Enterprise, adding real-time video avatars with lip-syncing and tool calling for customer service and interactive kiosks.
Google made Gemini 3.8 Live with Live Avatar generally available in Gemini Enterprise on September 24, adding near real-time generated video to its native voice models. Enterprises get video avatars with synchronized lip-syncing, fluid turn-taking, and asynchronous tool calling.
Customers deploy from a library of curated pre-built avatars, while custom avatars are gated behind a strict enterprise verification process to prevent misuse. Every generated audio and video stream carries an invisible SynthID watermark, and the service ships with US and EU endpoints.
Key Capabilities:

• Talking avatars: Real-time video with synchronized lip-syncing added to voice conversations.
• Misuse prevention: Custom avatars require strict enterprise verification; SynthID watermarks every output.
• Enterprise-ready: US and EU endpoints with provisioned throughput and compliance controls.

Google added local model support to the Antigravity SDK, letting developers run Gemma 4 26B directly on their own hardware for zero API costs, total data privacy, and no internet connection required.
Google updated the Antigravity SDK on September 23 to support local models, with initial support for Gemma 4 26B running through Google AI Edge's LiteRT. Developers can now execute agentic coding workflows entirely offline on their own GPU, with zero API costs.
A hybrid demo shows Gemini 3.8 Flash acting as cloud planner while local Gemma 4 instances do the heavy lifting, patching vulnerable code without any source ever leaving the machine. Over 97% of tokens in that workflow were processed locally.
Key Capabilities:

• Fully offline: Runs Gemma 4 26B on a local GPU with no internet required after setup.
• Zero API cost: Local workflows skip cloud inference charges and rate limits entirely.
• Hybrid option: A cloud model can plan while local Gemma instances execute, keeping code private.

Google launched Gemini 3.8 Flash TTS and Flash-Lite TTS, letting developers create custom voices from a text prompt or clone one from a 30-second sample, with line-by-line control over pacing and emotion.
Google launched Gemini 3.8 Flash TTS and Flash-Lite TTS on September 23, its most expressive audio models yet. Developers can design new voices from a natural-language prompt, pick from over 2,000 ready-made options, or clone a voice from a 30-second sample.
Flash TTS targets high-fidelity creative work like gaming and audiobooks with full line-by-line direction; Flash-Lite is built for cost-efficient scale, adjusting tone automatically for real-time voice agents. Both include built-in watermarking, and Flash TTS now tops Hume AI's voice benchmark.
Key Capabilities:

• Voice design from text: Describe a voice in plain language and get a custom character voice.
• Line-by-line direction: Control pacing, emotion, and dialect shifts for each line individually.
• Built-in safety: Watermarking included, plus a mandatory consent check for voice cloning.