What’s New in AI

Airbnb announced it will launch in the first half of 2027 an AI agent that plans and books travel: it will assemble maps, photos and comparisons around a request, and also help change a booking or fix a problem.

Source: AIdapted

On November 6, Kakao released Kanana-2-v-decide, a model that does not write text: it returns decisions (probabilities, choices, scores) so AI agents can judge faster mid-task. It answers in under 0.04 seconds on a single graphics card, and commercial use is allowed.

Source: Seoul Economic Daily

On November 6, DigitalOcean launched Cuttlefish, a native iPhone app for its Managed Agents platform, which lets users supervise their AI agents from their phone. The company also announced a collaboration with Box.

Source: AI Agent Store

On November 5, Perplexity launched Teammates, persistent agents that take ownership of ongoing work — triage, tracking, research — instead of waiting for a question, and that talk to the team on Slack. They are reserved for Max subscribers ($200 a month), in research preview on the web for now.

Source: Perplexity

On November 5, Anthropic opened Personal Finance in Claude: Pro, Max and Ultra subscribers in the United States can link their bank accounts via Plaid and ask questions about their spending. Access is read-only — Claude cannot move or invest your money — and the rollout is gradual, in beta.

Source: PromptWises

On November 5, Google made Gemini Omni 1.2 Flash generally available in the Gemini API and its business tools: 15-second video clips, conversational editing and a price of $0.07 per second in 720p, about 30% less than the previous version.

Source: SXF Brief

On November 5, OpenAI introduced GPT-6 Cyber, a model specialized in vulnerability research and defensive security testing: it scores 48.5% on the ExploitGym benchmark, versus 32.1% for the previous model. Access is initially reserved for approved security teams, via the API and Codex.

Source: OpenAI

On November 5, Moonshot AI formally launched Kimi Hosted Agents, a managed service that runs long tasks and lets several agents collaborate, with more than 140 skills and 30 plugins. It is initially reserved for enterprise-verified users.

Source: Unite.AI

On November 4, OpenAI launched Images 2.6 in ChatGPT, Codex and its API: generation is reportedly 50% faster, with animated GIFs, image markup and SVG exports. An update that directly affects image creation in ChatGPT.

Source: OpenAI

On November 4, OpenAI announced the planned acquisition of Pierre Computer Company, to expand Codex, its agentic coding tool. Neither the price nor an integration timeline was disclosed.

Source: The IT Guys

According to a Talkdesk survey published on November 4 of more than 1,000 US shoppers, 74% plan to use an AI assistant for their holiday shopping, and 15% would let an agent pay after a review step. A strong signal for agentic commerce.

Source: Agile Brand Guide

On November 4, Tencent launched Hy4, its new 770-billion-parameter model, which understands images and can operate a computer through its graphical interface and command line. It is already available in Tencent products and via API, but its open weights have not been published yet.

Source: Emergent.sh

Apple has opened Siri AI to all developers: through App Intents, any app can now offer its content and actions to Siri. The feature is in beta in iOS 27.3 and should reach users within a few months.

Source: AIToolsRecap

On November 3, CrowdStrike announced an “AI Kill Switch” in Falcon Guardian: an authorized administrator can stop an agent that drifts from its mission, keep it from relaunching and cut its network access. The stop runs through the gateway first, with direct on-device termination to follow.

Source: CrowdStrike

Menlo Ventures' 2026 report, published on November 3, puts enterprise AI spending at $176.8 billion by September 2026, 5.4 times more than in 2025. On OpenRouter, agents generate 80% of tokens consumed, roughly seven times more than humans.

Source: Menlo Ventures