Hey Devs! This week brought a wave of agentic AI innovations. Major announcements included Google’s Gemma 4 models, new agent-ready models from Microsoft, and upgrades to Claude that link it to Microsoft 365 and Windows PCs. Several tools expanded the context windows and introduced efficient on-device AI models. Here are the highlights relevant to developers.
<hr>
<h3>Google releases Gemma 4 open models</h3>
Google introduced Gemma 4, a suite of open models spanning 2 billion (E2B) and 4 billion (E4B) parameter edge models, as well as larger 26 billion Mixture-of-Experts and 31 billion dense models. The models were designed for advanced reasoning, agentic workflows, and native function calling; they offer multimodal vision and audio support and long 128K-256K-token context windows.
Read more: <a href="https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/" target="_blank" rel="noopener nofollow">https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/</a>
<hr>
<h3>Microsoft launches MAI-Transcribe-1, MAI-Voice-1 and MAI-Image-2 models</h3>
Microsoft released three models optimized for productivity: MAI-Transcribe-1 transcribes speech in the top 25 languages with a 2.5× speed improvement over Azure’s previous offering and low word-error rates; MAI-Voice-1 generates realistic speech at 60 seconds of audio per second and allows custom voice creation with a few seconds of input; and MAI-Image-2 generates images quickly with natural lighting, accurate skin tones, and high-quality diagrams.
Read more: <a href="https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-mai-transcribe-1-mai-voice-1-and-mai-image-2-in-microsoft-foundry/4507787" target="_blank" rel="noopener nofollow">https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-mai-transcribe-1-mai-voice-1-and-mai-image-2-in-microsoft-foundry/4507787</a>
<hr>
<h3>PrismML introduces Bonsai 8B 1-bit model</h3>
PrismML released Bonsai 8B, a 1.15 GB 8-billion-parameter language model that uses 1-bit weights (±1 with a shared scale factor). This design makes Bonsai 8B 14× smaller, 8× faster, and 5× more energy-efficient than full-precision models while maintaining comparable performance.
Read more: <a href="https://prismml.com/" target="_blank" rel="noopener nofollow">https://prismml.com/</a>
<hr>
<h3>Alibaba’s Qwen 3.6 Plus preview</h3>
Qwen 3.6 Plus offers a 1-million-token context window and up to 65K output tokens, enabling the model to process entire codebases and documents. It supports always-on chain-of-thought reasoning, native function calling, and multimodal input, making it suitable for agentic coding and large-scale repository analysis.
Read more: <a href="https://qwen.ai/blog?id=qwen3.6" target="_blank" rel="noopener nofollow">https://qwen.ai/blog?id=qwen3.6</a>
<hr>
<h3>Cursor 3: an agent-centric IDE</h3>
Cursor’s blog announced Cursor 3, a new development environment that treats agents as first-class citizens. The Agents Window allows running multiple local or cloud agents across repositories and handing sessions off between environments. A built-in browser lets agents open local websites and a plugin marketplace enables integration with third-party tools.
Read more: <a href="https://cursor.com/blog/cursor-3" target="_blank" rel="noopener nofollow">https://cursor.com/blog/cursor-3</a>
<hr>
GitHub Copilot CLI 101: The Complete Guide to Every Slash Command
GitHub Copilot CLI brings AI directly into your terminal, and the slash commands are where the real power hides. I tested all of them, so you know exactly which ones are worth your time.
Watch on YouTube: <a href="https://youtu.be/p7LakGgyb8M?si=tPjBWQFED4OtfIfN" target="_blank" rel="noopener nofollow">GitHub Copilot CLI</a>
<hr>
See you next week! Cheers, proflead!

Helping devs stay ahead with WebDev, AI & productivity tools.

proflead

A developer-focused AI news roundup covering major releases this week. Google launched Gemma 4 open models with multimodal support and long context windows up to 256K tokens. Microsoft released MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 for productivity use cases. PrismML introduced Bonsai 8B, a 1-bit 1.15 GB model that is 14× smaller and 8× faster than full-precision equivalents. Alibaba previewed Qwen 3.6 Plus with a 1-million-token context window suited for agentic coding. Cursor 3 was announced as an agent-centric IDE with multi-agent support, a built-in browser, and a plugin marketplace. A GitHub Copilot CLI guide covering slash commands is also highlighted.

AI News for Developers #19: The Agentic Stack Is Real (Gemma 4, Bonsai 8B, Cursor 3)