AI NewsOctober 4, 2026

Supabase raised $150 million and acquired Turso, betting that AI coding agents, not human developers, will soon create most new databases on its platform.
Supabase announced both moves on October 2. The company says it now adds over 1 million users and 4 million new databases monthly, with roughly 70% of those created by AI agents rather than people. Turso's architecture is built to provision many small, isolated databases quickly, which fits that agent-driven pattern better than Supabase's existing Postgres setup alone.
Turso founder Glauber Costa joins as Head of Agentic Services, and the platform will keep running independently with a planned migration path into Supabase's broader ecosystem over time.
Key Capabilities:

• Scale signal: 70% of new databases on Supabase are now created by agents, not humans.
• Leadership move: Turso's founder becomes Supabase's Head of Agentic Services.
• Funding: $150 million round led by GIC, four months after a $500 million raise.

A new 64GB version of NVIDIA's DGX Spark costs $4,999, less than the 128GB model, but still 25% more than that larger model's original launch price a year ago.
NVIDIA announced the DGX Spark 64GB on October 2, launching October 23 through partners including Acer, Dell, and HP. It keeps the same GB10 Grace Blackwell chip as the pricier model and can run AI models up to 100 billion parameters on its own, or 200 billion by linking two units together.
The launch highlights a broader memory shortage: the original 128GB model has jumped from its $3,999 launch price to $6,950 today. NVIDIA is pitching its Sync Cluster Assistant as a way to scale up by connecting multiple units instead of buying one bigger machine.
Key Capabilities:

• Lower entry price: $4,999, versus $6,950 for the current 128GB model.
• Cluster option: Two units linked together pool to 128GB and handle larger models.
• Same core chip: Uses the identical GB10 Grace Blackwell Superchip as the full-size version.

Anthropic's "mods" let developers write small TypeScript files that hook into Claude Code's internals, changing its behavior, interface, or adding entirely new features.
Anthropic released mods in Claude Code version 2.1.287 on October 1. A mod can watch an internal event, like a tool call or UI redraw, and rewrite or intercept it. Developers can write one by hand or ask Claude to build it during a session, with hot reload available for quick testing.
Early examples include a multiplayer Doom connector and a real-time cartoon spinner. Anthropic says it already rebuilt some of its own built-in features as mods. Because mods run with the same machine access as Claude Code itself, Anthropic recommends installing only from trusted sources.
Key Capabilities:

• Write or ask Claude: Build a mod in TypeScript yourself, or have Claude write it live.
• Hooks into events: Can intercept tool calls, prompts, slash commands, and UI rendering.
• Shareable: Ships inside plugins, installable via /plugin or shared through the Claude directory.

Kolibri is a bilingual English-German model anyone can download and run on their own servers, aimed at banks, public agencies, and manufacturers that can't send data to the cloud.
Aleph Alpha released Kolibri on October 3 under the Apache 2.0 license, a permissive license that allows free commercial use. It's a mixture-of-experts model, meaning it has many specialized sub-networks but only activates about 3.5 billion of its 78 billion total parameters per question, keeping it cheaper to run than its full size suggests.
It was trained from scratch in Germany and Finland and supports context windows up to 1 million tokens (roughly 750,000 words). Aleph Alpha also trained it to say "I don't know" rather than guess when it lacks solid evidence, a feature aimed at high-stakes, regulated work.
Key Capabilities:

• Self-hosted: Runs entirely on a company's own hardware, no cloud API required.
• Trained to abstain: Built to admit uncertainty instead of guessing on unclear questions.
• European-built: Trained entirely on infrastructure in Germany and Finland, with 80% of German training data curated in-house.

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.

Maket 2.0 turns a rough idea or an existing floor plan into an editable design, letting users ask for an extra bedroom or a new layout and see it rendered.
Montreal-based Maket announced Maket 2.0 on October 2, combining floor plan generation, conversational editing, and 3D rendering into one platform. Users can start from scratch or upload an existing plan, then describe changes in plain language, like "make the kitchen bigger."
Maket's CEO says the AI now handles roughly 70-75% of early schematic work, with licensed professionals still required for structural review and code compliance. Plans start around $20-30/month for individual users.
Key Capabilities:

• Two starting points: Begin from a text description or upload an existing floor plan.
• Conversational editing: Type plain-language changes and watch the layout update.
• Human review still required: Final structural and code compliance checks stay with licensed professionals.

Claude Opus 5.5 and Sonnet 5.5 are now selectable in Google's Antigravity coding environment, but only for paying Google AI Pro and Ultra subscribers, not free or trial users.
Google added both Claude models to Antigravity's official model picker on October 2-3, confirmed by Google DeepMind's Varun Mohan. They sit alongside Gemini and GPT-OSS as paid-tier options, billed through a Google subscription rather than Anthropic's own pricing.
Older models, including Claude Opus 4.6, Sonnet 4.6, and GPT-OSS-120B, are scheduled for removal from Antigravity on November 2. Some users report tighter usage limits on the new models, though Google hasn't confirmed that officially.
Key Capabilities:

• Paid-tier only: Requires a non-trial Google AI Pro ($19.99/month) or Ultra subscription.
• Heavy adoption: One developer reported Opus 5.5 already drawing over 50% of their coding prompts.
• Legacy cutoff: Claude 4.6 models and GPT-OSS-120B leave Antigravity on November 2.

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.