Qwen Releases Qwen3-Coder-Next | Google Introduces Agentic Vision in Gemini 3 Flash

Here is your today’s AI Dev Brief from Marktechpost, covering core research, models, infrastructure tools, and applied updates for AI developers and researchers. Also, don’t forget to register for NVIDIA GTC 2026 event (In person/Virtual). NVIDIA has been supporting us to bring free and unlocked AI research and dev news content to you.

Qwen3-Coder-Next is an open-weight 80B Mixture-of-Experts coding model from the Qwen team, built on the Qwen3-Next-80B-A3B backbone and optimized for agentic coding and local deployment. It activates only 3B parameters per token using a hybrid stack of Gated DeltaNet, Gated Attention, and sparse MoE layers, and supports a 256K token context for repository-scale tasks. The model is “agentically trained” on large collections of executable tasks with reinforcement learning, which improves long-horizon behaviors such as planning edits, calling tools, running tests, and recovering from failures. Benchmarks show strong SWE-Bench Verified, SWE-Bench Pro, SWE-Bench Multilingual, Terminal-Bench 2.0, and Aider scores that are competitive with much larger MoE models. Qwen3-Coder-Next exposes OpenAI-compatible APIs via SGLang and vLLM, and also ships as GGUF quantizations for local llama.cpp setups under Apache-2.0..… Read the full analysis/article here.

Google has introduced Agentic Vision in Gemini 3 Flash, a new capability that transforms image analysis from a passive "static glance" into an active investigation through a "Think → Act → Observe" reasoning loop. By integrating multimodal reasoning with Python code execution, the model can now autonomously perform complex visual tasks—such as zooming into fine-grained details, drawing annotations to justify its findings, and executing visual math or plotting—which has led to a 5–10% performance boost across vision benchmarks. This update, available via the Gemini API and Google AI Studio, enables developers to build more transparent and accurate visual agents that can audit their own reasoning and ground their answers in verifiable visual evidence..… Read the full analysis/article here.

▶ [Open Source] Rogue: An Open-Source AI Agent Evaluator worth trying Codes & Examples

▶ A Coding Implementation to Train Safety-Critical Reinforcement Learning Agents Offline Using Conservative Q-Learning with d3rlpy and Fixed Historical Data Codes Tutorial

▶ How to Build Advanced Quantum Algorithms Using Qrisp with Grover Search, Quantum Phase Estimation, and QAOA Codes Tutorial

▶ How to Build Multi-Layered LLM Safety Filters to Defend Against Adaptive, Paraphrased, and Adversarial Prompt Attacks Codes Tutorial

150+ more open codes/notebooks here ➡️