OpenAI Unveils Sora 2 AI Video Generation Model

OpenAI has launched the Sora 2 AI video generation model, accompanied by an introductory video showcasing its capabilities. Sora 2 offers enhanced realism, improved video quality and consistency, integrated speech and sound effects with video, and the ability to use personal likeness for cameo videos. Further details on Sora 2 can be found in the article 'Sora 2 Brings Us Social Audio-Video Cameos.'

Additionally, OpenAI introduced the Sora iOS app powered by the new Sora 2 model, featuring a TikTok-style feed for discovering generated videos and in-app AI video creation. The 'Cameos' feature allows users to appear in generated videos after capturing their image likeness, leading to the most viral Sora 2 videos shared. Sora is currently invite-only, with codes being distributed in waves.

Anthropic released Claude Sonnet 4.5, claiming it as the best in coding, agentic AI, and computer usage, highlighting its strong reasoning and tool-use capabilities. Sonnet 4.5 achieves a SOTA coding benchmark result of 77.2% on SWE-bench Verified, as well as SOTA on 'using computers' tasks.

Zhipu AI announced GLM-4.6, an update on GLM-4.5, offering longer context (up to 200K tokens), improved coding and reasoning performance, and lower token consumption in applied tasks. This update targets more capable agentic workflows, with a 40.5% performance on Terminal-Bench.

DeepSeek released DeepSeek-V3.2-Experimental, introducing DeepSeek Sparse Attention (DSA) to reduce compute for long-context training and inference while targeting quality parity with prior models. This model is based on DeepSeek-V3.1-Terminus.