Suno has introduced its new AI music model v5.5 along with features designed to showcase musical individuality
By Daniel Mercer Mistral AI aims to undercut competitors on price in speech recognition with Voxtral Transcribe 2. The second generation of its speech-to-text models starts at $0.003 per minute and, according to Mistral, delivers higher accuracy than models such as GPT-4o mini Transcribe, Gemini 2.5 Flash, and Deepgram Nova. The model family includes two variants: Voxtral Mini Transcribe V2, designed for processing large audio files, and Voxtral Realtime, built for real-time applications with latency under 200 milliseconds. Voxtral Realtime, which costs twice as much, uses a dedicated streaming architecture that transcribes audio as it arrives, targeting use cases such as voice assistants, live captions, and call center analytics
By Daniel Mercer Meta has completed the pretraining of its new AI model, codenamed “Avocado,” according to an internal memo.
By Chris Borden Chinese developer Kuaishou has unveiled the third version of its video generation model, Kling AI.
By Chris Borden The Chinese company Kling has released its Kling Video Model 3.0. The new model is described as an “all-in-one creative engine” for multimodal content creation. Key features include improved consistency of characters and visual elements, video generation with 15-second clips and enhanced control, as well as customizable multi-shot sequences.
By Alex Rowland Alibaba has unveiled Qwen3-Coder-Next, a new open-weight AI model designed for coding agents and local development. The model was trained on 800,000 verifiable tasks executed in runnable environments. Despite its relatively compact design—80 billion parameters in total, with only 3 billion active parameters—it delivers strong results on SWE-Bench Pro, a benchmark for coding agents.
By Chris Borden China’s AI labs are racing to roll out new models ahead of the Lunar New Year. According to the South China Morning Post, Zhipu AI and Minimax—both of which recently listed on the Hong Kong Stock Exchange—plan to update their flagship models within the next two weeks.
By Chris Borden Google DeepMind equips its Gemini 3 Flash model with a new capability called “Agentic Vision.” The model is designed to actively investigate images rather than merely observe them passively — although the feature does not yet work fully automatically in all cases.
By Daniel Mercer The AI research institute Allen AI (Ai2) has released SERA, a family of open-source coding agents designed to be easily adapted to private codebases at low cost. The flagship model, SERA-32B, solves up to 54.2% of tasks on the SWE-Bench-Test Verified coding benchmark (with 64K context), outperforming comparable open-source models.
By Alex Rowland There Is a New Best Math Model. OpenAI’s GPT-5.2 Pro Sets a New Record on FrontierMath
ES 
EN