📰 Dev.to · albe_sf

Google's release of Gemma 2, in 9B and 27B parameter sizes, marks a significant step in open model performance, offering a competitive alternative to models twi

Xiaomi just open-sourced its MiMo-V2.6 model family. The real story isn't just the new weights, but their public bet on scaling reinforcement learning as the pa

Google just shipped Gemini 3.8 Live and Extended Thinking. The key takeaway for builders is the introduction of asynchronous, background reasoning, which requir

The general availability of Snowflake's Cortex Coding Agent signals a shift from generic code completion to platform-native, data-aware agentic workflows. Here'

Snowflake AI Research's Arctic-SnowCoder suggests that for code generation, a smaller, higher-quality dataset can outperform models trained on trillions of toke

Google bundled its Antigravity autonomous coding agent into Gemini Enterprise subscriptions. This isn't just a feature update; it's a fundamental shift in how a

OpenAI's new Computer History feature for macOS moves ChatGPT from a window into an OS-level observer. But its unencrypted local storage creates a new class of

Anthropic's new model isn't just an incremental update. Claude 3.5 Sonnet beats the flagship Opus on key benchmarks like coding, at twice the speed and a fracti

Snowflake's latest Cortex AI updates introduce serverless fine-tuning. This isn't just a feature drop; it's a structural shift in how enterprise AI gets built,

Google's PaliGemma models are powerful, open vision-language components. But they aren't for chat; they're designed for transfer learning. Here's how to use the

Model choice is being automated away inside the IDE. Cursor's new intelligent router and team-level controls are a bigger deal than they look, changing how you

The new Gemini 3.6 Flash and 3.5 Flash-Lite models are now GA. The real story isn't just speed; it's the deprecation of sampling parameters, forcing a new relia

Alibaba Cloud's release of the Qwen2 family of open-source models provides a credible, high-performing alternative to the usual suspects. With strong multilingu

Google just released Gemma 2, a new family of open models in 9B and 27B sizes. It's engineered for performance and efficiency, offering a serious alternative fo

Anthropic's latest model, Claude 3.5 Sonnet, is not just an incremental update. Its combination of Opus-level intelligence at twice the speed and a fraction of

OpenAI is shutting down 13 legacy model snapshots on July 23, including the entire pre-GPT-5.3 Codex series. This is not a 1:1 swap, and your systems will break

Mistral AI has released Robostral Navigate, an 8B model for robot navigation that uses only a single RGB camera. This is a significant step in lowering the hard

Anthropic's new Artifacts feature in Claude 3.5 Sonnet is more than a preview pane. It's a fundamental shift in the developer workflow, moving us from conversat

xAI's latest model, Grok 4.5, was released on July 8. It's not just another benchmark chaser; its real story is token efficiency and deep integration with devel

Google's new Gemma 2 open models are available in 9B and 27B sizes. The real story isn't just performance, but the subtle architectural changes that impact effi

Cohere's new Aya 23 family of open-weight models provides a strong new baseline for multilingual applications. We'll look at what the 8B and 35B parameter model

Snowflake announced Cortex Sense, a system that automatically builds a semantic model of your business from existing queries and tools. This is a direct attempt

Most generative video models ship silent. Google DeepMind's new V2A technology generates synchronized audio from video pixels and text prompts, pointing to a fu

Cohere's Compass model tackles multi-aspect data, a core reason why retrieval-augmented generation systems fail on complex enterprise documents. It’s a move awa