Neuro Weekly Update | AI Model Advancements and New Features

Neuro Review for the Past Week (#109)

LLM
– Google’s multimodal embeddings — Gemini Embedding now natively supports videos up to 2 minutes, as well as PDFs, audio, and text files.
– In Claude, the maximum context length has been increased by default to one million tokens — the additional charge for long queries in Sonnet/Opus 4.6 has been removed. Additionally, the limits for off-peak hours have been doubled for two more weeks.

Generative Models
– 2D image processing directly in Photoshop — Adobe has implemented a highly anticipated feature that significantly simplifies working with image compositions. Creating complex collages is now even easier.

Miscellaneous
– The new version of Flash Attention 4 has been released — optimized for Blackwell and providing up to 1.3x acceleration when working with BF16. All software is now fully written in Python using CuTe-DSL.

Created with n8n:
https://cutt.ly/n8n

Created with syllaby:
https://cutt.ly/syllaby

Page view /ai-blog/humanoid-robot-protoclone-polish-innovation-in-robotics 17.03 19:29 Page view /ai-blog/gpt-5-4-mini-nano-fast-efficient-ai-models-openai/ 17.03 19:28 Page view 17.03 19:28 Page view 17.03 19:26 Page view 17.03 19:25 Page view 17.03 19:23 Page view /ai-blog/us-iran-military-situation-update-trump-aims-for-resolution/ 17.03 19:18 Page view 17.03 19:18 Page view /ai-blog/kling-3-0-ai-creativity-platform-seamless-multimedia-content/ 17.03 19:17 Page view /ai-blog/un-human-rights-office-warns-of-survival-crisis-latest-updates/ 17.03 19:16