What\u2019s new in AI tools
Automated weekly roundup of release notes and changelogs across every AI tool we track. Currently indexing 317 updates.
This week \u2014 2026-W18
AI Tools This Week: 42 updates across 7 products
-
Groq Compound and Compound Mini
Production-ready agentic AI systems moved from beta to general availability, integrating web search, code execution, and browser automation in a single API call. Built on GPT-OSS-120B and Llama models.
-
Moonshot AI Kimi K2 Instruct 0905
New model release with 256K context window, prompt caching support, and enhanced agentic coding capabilities. Offers 200+ t/s performance at $1.50/M tokens blended pricing.
-
Remote Model Context Protocol (MCP)
Beta release of Remote MCP server integration on GroqCloud, enabling AI models to connect to thousands of external tools through Anthropic's open MCP standard. Fully compatible with OpenAI APIs for seamless migration.
-
Introducing Forge
Forge is a new system that allows enterprises to build frontier-grade AI models grounded in their proprietary knowledge.
-
Speaking of Voxtral
Voxtral TTS is a new open-weights text-to-speech model that produces fast, adaptable, and lifelike speech for voice agents.
-
Connect the dots: Build with built-in and custom MCPs in Studio
New connectors in Studio allow enterprises to connect data to AI applications with reusable connectors, tool calling, and approval controls.
-
Remote agents in Vibe. Powered by Mistral Medium 3.5.
Mistral AI introduced Mistral Medium 3.5 model with remote coding agents in Vibe and a new Work mode in Le Chat for complex tasks.
-
Announcing the Databricks analytics engineer learning pathway
Databricks launched a new Analytics Engineer Learning Pathway to provide structured training for analytics engineers.
-
Expanded interoperability with Unity Catalog Open APIs
Unity Catalog introduces expanded Open APIs to improve interoperability across different data platforms and tools.
-
AI QA Analyst, Custom Alert Notifications, Node Search, and everything else from this month
Introduced AI QA Analyst for automatic call review and feedback, custom alerting system for monitoring agent performance, and conversation flow node search functionality.
-
Agent Guardrails, Webhook Testing, Auto-Failover, Billing Transparency & more
Added safety guardrails to block harmful content and jailbreaks, webhook testing tools, global node return paths, and billing transparency improvements with separate TTS line items.
-
ChatGPT app, Dynamic voice speed, A/B testing, and more
Retell AI launched a ChatGPT app for building voice agents, dynamic voice speed that adapts to caller pace, and A/B testing capabilities for splitting call traffic across multiple agents.
-
Python SDK v0.31.1, TypeScript SDK v0.32.0 changed
Updated SDKs with improved chat completion message type definitions for better OpenAI compatibility and added support for new Groq Compound tools including Wolfram Alpha and Browser Automation.
-
Groq Compound and Compound Mini added
Compound and Compound Mini are Groq's production-ready agentic AI systems that integrate web search, code execution, and browser automation into a single API call. Moving from beta to general availability, these systems deliver ~25% higher accuracy and ~50% fewer mistakes across benchmarks.
-
Moonshot AI Kimi K2 Instruct 0905 added
Kimi K2-0905 brings Moonshot AI's cutting-edge model to GroqCloud with enhanced agentic coding capabilities and a 256K context window. The model delivers improved frontend development performance and includes prompt caching for up to 50% cost savings.
-
Remote Model Context Protocol (MCP) added
Remote Model Context Protocol (MCP) server integration is now available in Beta on GroqCloud, connecting AI models to thousands of external tools through Anthropic's open MCP standard. Developers can connect any remote MCP server to models hosted on GroqCloud with zero code changes from OpenAI.
-
3.5: Notion Developer Platform
Launch of the Notion Developer Platform enabling developers to sync any data source, build custom agent tools, and orchestrate external agents within Notion.
-
EvenUp Introduces a New Era of PI Operations with Pre-Litigation as a Service (PLAAS)
EvenUp launched Pre-Litigation as a Service (PLAAS), combining purpose-built AI with expert case managers to help personal injury firms scale quality outcomes.
-
3.5: Notion Developer Platform
Launched the Notion Developer Platform with Workers runtime, external agents API, database sync capabilities, and CLI for developers and coding agents.
-
EvenUp Introduces a New Era of PI Operations with Pre-Litigation as a Service (PLAAS)
EvenUp launches Pre-Litigation as a Service (PLAAS), combining purpose-built AI with expert case managers to help personal injury firms scale quality outcomes.
-
3.5: Notion Developer Platform
Notion launched a comprehensive developer platform allowing teams to sync any data source, build custom agent tools, and orchestrate external agents like Claude and Codex. The platform includes Workers for hosted code execution, a CLI for developers, and APIs for database sync and webhook triggers.
-
3.5: Notion Developer Platform
Notion launched a comprehensive developer platform allowing users to sync any data source, build custom agent tools, and orchestrate external agents like Claude and Codex. The platform includes Workers for custom code execution, CLI tools, and APIs for extending Notion without managing infrastructure.
-
EvenUp Introduces a New Era of PI Operations with Pre-Litigation as a Service (PLAAS)
EvenUp launches Pre-Litigation as a Service (PLAAS), combining purpose-built AI with expert case managers to help firms scale quality outcomes.
-
Helix-02 Bedroom Tidy
The Helix-02 system demonstrates autonomous bedroom cleaning and organization capabilities.
-
Helix-02 Bedroom Tidy
Figure AI demonstrated Helix-02's ability to perform bedroom cleaning and organization tasks autonomously.
-
Helix-02 Bedroom Tidy
Figure AI's Helix-02 robot demonstrates autonomous bedroom cleaning and organization capabilities.
-
Pushing the Frontier for Data Agents with Genie
Databricks enhanced Genie, their state-of-the-art data agent designed for answering complex questions. The improvements advance the capabilities of data agents in handling sophisticated analytical queries.
-
MCP Marketplace brings real-time intelligence to agentic applications
Databricks launched MCP Marketplace to provide real-time intelligence capabilities for agentic AI applications. The marketplace enables AI systems to access business context and reasoning capabilities.
-
Using MemAlign to Improve Evaluation of Traditional Machine Learning in Genie Code
Databricks introduced MemAlign to enhance evaluation of traditional machine learning within Genie Code, their autonomous AI partner. This improvement focuses on better assessment capabilities for ML models.
-
How Superhuman and Databricks built a 200K QPS inference platform together
Superhuman partnered with Databricks to build a high-performance inference platform capable of handling 200,000 queries per second. The collaboration demonstrates real-time AI inference capabilities at scale.
-
Plan Mode
Introduced Plan Mode for Custom Agents, which adds a preliminary step where agents ask clarifying questions and build detailed plans before executing complex multi-step tasks. This reduces errors and increases confidence in agent outputs.
-
Plan Mode
Introduced Plan Mode where agents ask clarifying questions and build detailed plans before executing complex multi-step tasks. This feature reduces surprises and increases confidence when agents make significant changes to pages or databases.
-
How AI Communication Agents Expanded Operational Capacity 2.5x in the First 90 Days
EvenUp's AI Communication Agents demonstrated significant operational leverage by expanding capacity 2.5x within 90 days of launch.
-
New Custom Agent Directory
Added a dedicated Custom Agent Directory in the Library where users can browse all workspace agents, pin favorites, and create new ones. Agents are now accessible via Library → Agents in the sidebar.
-
How AI Communication Agents Expanded Operational Capacity 2.5x in the First 90 Days
EvenUp's AI Communication Agents proved operational leverage within 90 days post-launch, expanding capacity by 2.5x.
-
New Custom Agent Directory
Added a dedicated Custom Agent Directory in the Library where users can browse all workspace agents, pin favorites, and create new ones. The directory provides centralized access to automate team workflows.
-
ChatGPT app, Dynamic voice speed, A/B testing, and more
Retell AI launched a ChatGPT app for building voice agents, dynamic voice speed matching caller pace, and A/B testing for splitting call traffic across multiple agents.
-
Python SDK v0.31.1, TypeScript SDK v0.32.0
Updated SDKs with improved chat completion message type definitions for better OpenAI compatibility and added support for new Groq Compound tools including Wolfram Alpha and Browser Automation.
-
Groq Compound and Compound Mini
Compound and Compound Mini are Groq's production-ready agentic AI systems that integrate web search, code execution, and browser automation into a single API call. Moving from beta to general availability, these systems deliver ~25% higher accuracy and ~50% fewer mistakes across benchmarks.
-
Moonshot AI Kimi K2 Instruct 0905
Kimi K2-0905 brings Moonshot AI's cutting-edge model to GroqCloud with day zero support, featuring a 256K context window and enhanced agentic coding capabilities. The model offers prompt caching with up to 50% cost savings and runs at 200+ t/s at $1.50/M tokens blended pricing.
-
Remote Model Context Protocol (MCP)
Remote Model Context Protocol (MCP) server integration is now available in Beta on GroqCloud, connecting AI models to thousands of external tools through Anthropic's open MCP standard. Developers can connect any remote MCP server to models hosted on GroqCloud with zero code changes from OpenAI.
-
New Custom Agent controls for admins
Added comprehensive admin controls for Custom Agents including permission management, per-agent credit limits, workspace-level spending controls, and usage tracking dashboards. Includes automatic guardrails to prevent unexpected spending.
-
New Custom Agent controls for admins
Released admin controls for Custom Agents including permission management, per-agent credit limits, workspace-level spending controls, and usage tracking dashboards. Includes automatic guardrails to prevent unexpected spending.
-
A new home on mobile
Updated the mobile app with a redesigned home experience for improved navigation and usability on mobile devices.
-
A new home on mobile
Updated the mobile app with a new home screen interface to improve navigation and user experience on mobile devices.
-
Ramping Figure 03 Production
Figure AI announced the scaling up of production for their Figure 03 humanoid robot model.
-
Remote agents in Vibe. Powered by Mistral Medium 3.5.
Mistral AI introduces Mistral Medium 3.5 model, remote coding agents in Vibe, and a new Work mode in Le Chat for handling complex tasks.
-
Ramping Figure 03 Production
Figure AI scales up manufacturing of their Figure 03 humanoid robot model.
-
Ramping Figure 03 Production
Figure AI is scaling up production of their Figure 03 humanoid robot model.
-
Remote agents in Vibe. Powered by Mistral Medium 3.5.
Mistral AI introduced Mistral Medium 3.5 with remote coding agents in Vibe and a new Work mode in Le Chat for complex tasks.