# AI News Digest - 2025-12-22

1. Nvidia released NitroGen, a vision-action base model that had been trained on about 40,000 hours of gameplay from over 1,000 games to serve as a foundation for creating universal AI agents across game worlds.

2. Anthropic's Claude Opus 4.5 achieved a 50 percent time horizon of approximately 4 hours and 49 minutes in METR tests, demonstrating the model could sustain some tasks lasting nearly five hours.

3. Alibaba's Qwen unit released Qwen-Image-Layered, an image-editing model that split images into multiple editable RGBA layers with transparent backgrounds for layer-wise editing similar to Photoshop.

4. OpenAI added tone controls to ChatGPT, introducing new personalization options that allowed users to customize how the assistant communicated.

5. Kuaishou released Kling 2.6, which added voice control and motion upgrades to its AI video generation tools to advance realism in generated video.

# References

1. [https://the-decoder.com/nvidia-wants-to-create-universal-ai-agents-for-all-worlds-with-nitrogen/](https://the-decoder.com/nvidia-wants-to-create-universal-ai-agents-for-all-worlds-with-nitrogen/)

[Nvidia wants to create universal AI agents for all worlds with NitroGen](https://the-decoder.com/nvidia-wants-to-create-universal-ai-agents-for-all-worlds-with-nitrogen/)

1. [https://the-decoder.com/anthropics-claude-opus-4-5-can-tackle-some-tasks-lasting-nearly-five-hours/](https://the-decoder.com/anthropics-claude-opus-4-5-can-tackle-some-tasks-lasting-nearly-five-hours/)

[Anthropic's Claude Opus 4.5 can tackle some tasks lasting nearly five hours](https://the-decoder.com/anthropics-claude-opus-4-5-can-tackle-some-tasks-lasting-nearly-five-hours/)

1. [https://the-decoder.com/alibabas-qwen-releases-ai-model-that-splits-images-into-editable-layers-like-photoshop/](https://the-decoder.com/alibabas-qwen-releases-ai-model-that-splits-images-into-editable-layers-like-photoshop/)

[Alibaba's Qwen releases AI model that splits images into editable layers like Photoshop](https://the-decoder.com/alibabas-qwen-releases-ai-model-that-splits-images-into-editable-layers-like-photoshop/)

1. [https://the-decoder.com/chatgpt-gets-tone-controls-openai-adds-new-personalization-options/](https://the-decoder.com/chatgpt-gets-tone-controls-openai-adds-new-personalization-options/)

[ChatGPT gets tone controls: OpenAI adds new personalization options](https://the-decoder.com/chatgpt-gets-tone-controls-openai-adds-new-personalization-options/)

1. [https://the-decoder.com/kling-2-6-adds-voice-control-and-motion-upgrades-as-ai-video-tools-race-toward-realism/](https://the-decoder.com/kling-2-6-adds-voice-control-and-motion-upgrades-as-ai-video-tools-race-toward-realism/)

[Kling 2.6 adds voice control and motion upgrades as AI video tools race toward realism](https://the-decoder.com/kling-2-6-adds-voice-control-and-motion-upgrades-as-ai-video-tools-race-toward-realism/)

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