Weekly digest · week 43
What's new in AI tools: week of Oct 19 – 25, 2026
Short answer: 13 dated updates from Kimi, Canva, Sloosh and 9 more, each linked to its source; 7 tools added to the directory.
Releases & updates
Dated launches and product changes, each linked to its source.
Kimi — Kimi K3.1 ProMoonshot introduced K3.1 Pro, the first K3.1-series model and its most capable 3T-class open-weight model so far, with a 1M-token context window and gains in coding, 3D creation, financial research and agent tasks. A K3.1 Pro Highspeed variant runs 2-3x faster, and full weights are promised by Nov 4. Source
Canva — Canva Craft image modelCanva Research released Craft, Canva's first in-house image generation model, built for design work and trained on human-made Canva templates. It plans the composition before rendering, generates images up to 2K in more than 30 aspect ratios, renders text in 55 languages, and works with Magic Layers to turn results into editable layers. It's available to all users in Dream Lab and to developers through the preview image generation API (craft_v1_2k), using AI credits. Source
Sloosh — Sloosh launches globallyShopOS opened Sloosh to creators and brand teams worldwide, built on the same image and video engine it uses for brand work, with an MCP server for ChatGPT, Claude, Gemini and Cursor and an After Effects plugin announced. Source
Breken — Breken launches with the ESCAPES benchmarkBreken launched publicly alongside ESCAPES, an open benchmark of 327 user-reported bugs from 33 open-source products; in Breken's own vendor-run test it caught 76 of the 235 bugs present in July 1 code (32%). Source
Claude — Claude Duet research preview (proactive agent app)Anthropic opened a waitlist research preview of Claude Duet, a standalone iOS, Android and claude.ai app where Claude works as a proactive agent: a morning brief across connected apps, background tasks on its own cloud computer, and approval before it sends, pays, signs or deletes. US Pro and Max plans only, using the same usage limits. Source
Beacon — Asymptote Labs raises $5.5M seed to build on BeaconTower Research Ventures led an oversubscribed $5.5M seed round; Beacon stays MIT-licensed on GitHub, with a paid enterprise platform for agent runtime security built on top. Source
Gemini — New Gemini tools for small business ownersGoogle added a business landing page in the Gemini app, brand-kit uploads for on-brand visuals with Nano Banana 2.1, financial insights from spreadsheets and receipts, and connections to Google Business Profile, Google Ads, Workspace and apps such as Canva, Shopify and Xero. Source
ElevenLabs — Eleven Realtime speech-to-speech modelsElevenLabs began rolling out Eleven Realtime, a family of speech-to-speech models for ElevenAgents voice agents: v1 Mini (the default, fastest and cheapest) and v1 Max (reasons before answering, for complex instructions and multi-step tool calls). Clients built on OpenAI's Realtime API can switch by changing the endpoint and session config. Source
Gemini — Gemini API CLI (experimental preview)Google released gemini-api, an open-source (Apache 2.0) command-line tool that lets developers and shell-capable coding agents call Gemini text, image, video, music and speech models, analyze files and run managed agents like Deep Research. It is labeled an experimental preview. Source
Harmony — Launch for sales teamsmonday.com launched Harmony, a voice AI agent platform for sales teams built on task-specific models, with an AI dialer, inbound routing, compliance controls and CRM write-back. Source
Notion AI — Rebuilt Notion Agents and a cheaper Auto model routerNotion Agents now get their own computer and browser, a "Review for me" mode with safety checks and tasks of up to 24 hours. The upgraded Auto router picks a model per task at about 90% lower average cost, and a new credit system and an improved Notion MCP launched alongside. Source
Wabi — Wabi 2.0 opens to all usersWabi relaunched as a shared personal AI agent: multiple task threads, AI-built apps inside each thread, and threads that friends or family can join, with data access scoped per chat. Source
Zendesk AI — Employee Service AI Agents (early access)Zendesk introduced Employee Service AI Agents and Advanced Action Flows that resolve IT, HR and workplace requests end to end, such as access approvals and onboarding. They are in open early access, with general availability expected in the coming weeks. Source
Added to Pickfinch
New in the directory this week, best score first. Ranks are within each tool's main use case.