# AI News Digest - 2026-04-03

1. Google released Gemma 4, its most capable open model family, and distributed four new models under the Apache 2.0 license for the first time, with variants optimized for devices ranging from smartphones to workstations.

2. Microsoft introduced MAI-Transcribe-1, a speech-to-text model that ran 2.5× faster than its predecessor, processed 25 languages with robustness to background noise, and was priced at about $0.36 per audio hour while being deployed in Microsoft products.

3. Nvidia set new MLPerf inference records using 288 GPUs in the latest benchmark round, which for the first time included multimodal and video models, while AMD and Intel emphasized different performance metrics.

4. Alibaba launched Qwen3.6-Plus, marking its third proprietary AI model release in a matter of days.

5. Nvidia, UC Berkeley and Stanford released a framework showing that state-of-the-art AI models failed at robot control without human-designed building blocks, and demonstrated that agentic scaffolding techniques, including targeted test-time compute scaling, substantially closed the performance gap.

# References

1. [https://the-decoder.com/googles-gemma-4-is-now-available-with-apache-2-0-licensing-for-the-first-time/](https://the-decoder.com/googles-gemma-4-is-now-available-with-apache-2-0-licensing-for-the-first-time/)

[Google's Gemma 4 is now available with Apache 2.0 licensing for the first time](https://the-decoder.com/googles-gemma-4-is-now-available-with-apache-2-0-licensing-for-the-first-time/)

1. [https://the-decoder.com/microsofts-mai-transcribe-1-runs-2-5x-faster-than-its-predecessor-at-0-36-per-audio-hour/](https://the-decoder.com/microsofts-mai-transcribe-1-runs-2-5x-faster-than-its-predecessor-at-0-36-per-audio-hour/)

[Microsoft's MAI-Transcribe-1 runs 2.5x faster than its predecessor at $0.36 per audio hour](https://the-decoder.com/microsofts-mai-transcribe-1-runs-2-5x-faster-than-its-predecessor-at-0-36-per-audio-hour/)

1. [https://the-decoder.com/nvidia-sets-new-mlperf-records-with-288-gpus-while-amd-and-intel-focus-on-different-battles/](https://the-decoder.com/nvidia-sets-new-mlperf-records-with-288-gpus-while-amd-and-intel-focus-on-different-battles/)

[Nvidia sets new MLPerf records with 288 GPUs while AMD and Intel focus on different battles](https://the-decoder.com/nvidia-sets-new-mlperf-records-with-288-gpus-while-amd-and-intel-focus-on-different-battles/)

1. [https://the-decoder.com/alibaba-launches-qwen3-6-plus-its-third-proprietary-ai-model-in-days/](https://the-decoder.com/alibaba-launches-qwen3-6-plus-its-third-proprietary-ai-model-in-days/)

[Alibaba launches Qwen3.6-Plus, its third proprietary AI model in days](https://the-decoder.com/alibaba-launches-qwen3-6-plus-its-third-proprietary-ai-model-in-days/)

1. [https://the-decoder.com/ai-models-fail-at-robot-control-without-human-designed-building-blocks-but-agentic-scaffolding-closes-the-gap/](https://the-decoder.com/ai-models-fail-at-robot-control-without-human-designed-building-blocks-but-agentic-scaffolding-closes-the-gap/)

[AI models fail at robot control without human-designed building blocks but agentic scaffolding closes the gap](https://the-decoder.com/ai-models-fail-at-robot-control-without-human-designed-building-blocks-but-agentic-scaffolding-closes-the-gap/)

For the site tree, see the [root Markdown](https://slashpage.com/ixtj-dev.md).