JetBrains

Building AI Agents with Kotlin ADK

This tutorial demonstrates a starter "Hello World" AI agent built with the Kotlin Agent Development Kit (ADK). The sample project (on GitHub) shows a Kotlin LlmAgent configured to use Google's Gemini model and to discover and call a local "greet" tool via a Model Context Protocol (MCP) Ktor server. The repo includes two Gradle modules (agent and server), uses Kotlin 2.3.0, Kotlin ADK SDK v0.6.0, MCP Kotlin SDK v0.8.1, Ktor 3.0.0, Java 25, and Gradle 9.2.1. The tutorial covers local runs, unit tests, an MCP smoke test that does not require a Gemini API key, and instructions to containerize and deploy the MCP server to Google Cloud Run.

Three Juejin AI roundups use different scorecards

The author compared three late-2025 Juejin roundup posts that each purported to answer "which AI to use in 2026" but used different evaluation frameworks. One December piece used a five-axis decimal scorecard (CodeBuddy top at 9.6, Blackbox bottom at 7.2). A year-end "yuan-mode" review labeled Gemini the first choice for multimodal and ultra-long context, ChatGPT the first choice for generality, and Claude for code—listing identical monthly prices (140元) for Gemini Pro, ChatGPT Plus, and Claude Pro. A November frontend-focused roundup rated Cursor best overall, GitHub Copilot best for ecosystem integration, Codeium best value, and V0.dev for frontend UI with five-star ratings. The author argues the three formats answer different reader jobs, forcing engineers to translate between scorecards manually and notes their personal multi-tool stack and plans to reassess in three months.

JetBrains Releases Mellum2 12B Mixture-of-Experts Model

JetBrains announced Mellum2, an open-source 12-billion-parameter model based on a Mixture-of-Experts (MoE) architecture released June 1, 2026. Mellum2 activates only ~2.5 billion parameters per inference, which the team says yields inference speeds over twice as fast than equivalent-scale models and reduces deployment costs. The model is optimized for text and code (no multimodal inputs), positioned as a "focused" model for multi-model collaboration systems handling tasks like prompt classification, tool selection, context compression for RAG pipelines, sub-agent planning validation, and code completion. Mellum2 is released under the Apache 2.0 license; a technical report is on arXiv (ID 2605.31268) and model weights are available on HuggingFace. Benchmarks reportedly show competitive performance among open-source models of similar scale.