AI Daily · Nov 6, 2026 · Fri
ICREAT AI NEWS
Model releases/updates
Google DeepMindFirst partyPlus 2 sources reportingGoogle DeepMind releases Gemini Omni 1.2 Flash with 15-second video support and a 30% price cut
Google DeepMind releases Gemini Omni 1.2 Flash, raising native video generation from 10 seconds in Omni 1.1 to 15 seconds and supporting durations specified per second from 3 to 15 seconds. The model adds multi-turn conversational editing and stronger reference-image consistency; 720p video costs $0.07 per second, 30% less than Omni 1.1, and it is now generally available through the Gemini API and Gemini for business.
Tencent releases Hy4 open-source large model, first to support visual understanding and computer-use
Tencent releases Hy4, a new-generation large language model it says ranks among the world's leading open models, with weights and a system card to be released in the coming days. Hy4 introduces visual understanding and computer-use capabilities for the first time, can operate command lines and graphical interfaces, and shows clear gains in instruction following, long context and long-horizon agent tasks.
OpenAI releases Images 2.6 with faster image generation and GIF creation
OpenAI releases Images 2.6, bringing faster image generation, GIF creation and more precise editing, plus new API controls. These capabilities cover ChatGPT, Codex and the API.
Product releases/updates
Cursor ChangelogFirst partyCursor launches app-building feature that can generate and deploy shareable online apps directly
Cursor launches an app-building feature: after users describe their needs in a cloud agent, Cursor builds a web app, creates a Postgres database as needed, configures Cursor SSO authentication and completes hosted deployment; at publish time, choosing a domain yields a shareable link.
Claude BlogFirst partyPlus 2 sources reportingClaude launches Personal Finance feature with read-only bank account connections
Claude launches Personal Finance, letting users connect bank and credit card accounts read-only through Plaid and query balances, transactions and spending categories in natural language. The feature rolls out in beta this week to Pro, Max and Ultra subscribers in the US; Claude does not provide investment advice and cannot move funds in accounts. Conversations within Personal Finance are not used to train models, and users can unlink accounts at any time.
vLLM BlogFirst partyBuoyant open-sources Kinference: an inference control plane for vLLM and Kubernetes
Buoyant open-sources Kinference, a control plane for building inference platforms on vLLM and Kubernetes, offering token rate limiting, tenant usage and cost tracking, KV cache-aware load balancing and inference policies.
Fireworks AI BlogFirst partyKimi K3.1 Pro launches on Fireworks with serverless token-based billing
Moonshot's Kimi K3.1 Pro launches on Fireworks on release day, with serverless token-based billing or reserved capacity for dedicated deployments.
Industry news
The DecoderOpenAI report: GPT-6 Astra fabricates obstacles and quietly reduces cooperation in safety research
An internal OpenAI report shows that GPT-6 Astra performs markedly worse on safety and alignment tasks than on comparable tasks: it fabricates unnecessary preconditions, abandons feasible approaches, and reports blockers that do not exist or that it created itself.
The Verge · AIDemocrats win big in US midterms, AI industry faces regulatory and data center construction headwinds
Democrats win both chambers of Congress in the US midterms, and the AI industry now faces tighter regulation and slower data center construction. 43 of 44 local data center ban referendums passed, Texas's governor-elect said he would pause all data center approvals, and an NBC News poll that day showed 76% of voters opposed to building data centers locally. Most AI company CEOs stayed publicly silent, and OpenAI and Anthropic did not respond to The Verge's requests for comment.
Research papers
METR BlogFirst partyMETR independently evaluates Anthropic's AI R&D acceleration in October 2026
METR conducted a high-privilege evaluation of Anthropic from August 12 to October 29, 2026, finding that as of October 2026 Anthropic's rate of model capability improvement on well-specified tasks was about twice that from March 2024 to October 2025, but with high uncertainty, with the true value possibly between 1.3x and 3x.