AI NewsAugust 29, 2026

 Gemini Omni 1.1 Flash keeps characters, lighting, and story flow consistent across extensions.
Google launched Gemini Omni 1.1 Flash, letting users analyze up to 10 seconds of footage to generate 10-second continuations, building clips up to 40 seconds total. First-and-last frame controls keep transitions smooth.
Quick 360p drafts let creators test ideas before upscaling to 1080p or 4K. It tops the Text-to-Video Arena at 1,515 points, with pricing starting at $0.03 per second.
Key Capabilities:

• Scene Extension: Builds clips up to 40 seconds by chaining 10-second video continuations together.
• Fast Drafting: Generates quick 360p test versions before upscaling to full 1080p or 4K quality.
• Leaderboard Topper: Scores 1,515 points on the Text-to-Video Arena, leading current rankings.

 It claimed the #1 spot in the Video Edit Arena, edging out rivals by a slim margin.
Alibaba's Wan 3.0 claimed the top spot in the Video Edit Arena with 1,414 points, narrowly leading Dreamina Seedance 2.5 by 4 points and MiniMax H3 by 22. It's Alibaba's debut in this specific arena.
The Video Edit Arena now tracks 10 models from 8 labs competing for video editing supremacy. Alibaba's entry shows the space is getting crowded fast.
Key Capabilities:

• Top Ranked: Scored 1,414 points, narrowly beating Dreamina Seedance 2.5 for the #1 spot.
• Close Competition: Leads MiniMax H3 by just 22 points in a tightly packed leaderboard.
• Growing Field: The arena now features 10 models from 8 different AI labs.

Expert Intelligence lets you combine trusted published sources directly into your research notebook.
Google introduced Expert Intelligence in Gemini Notebook, a cross-Google initiative starting with eligible Google Play ebooks. Users can now combine expertise from favorite authors with other sources they already use.
This turns Gemini Notebook into a place where published book knowledge blends with your own notes and research. It's a step toward AI tools grounded in trusted, licensed material.
Key Capabilities:

• Book Integration: Pulls in eligible Google Play ebooks as trusted sources inside your notebook.
• Author Expertise: Lets users combine insights from favorite authors alongside their own research sources.
• Cross-Google Feature: Built as a joint initiative connecting Google Play and Gemini Notebook.

 The staged rollout is over, with Fast, Expert, Heavy, Auto, and Build all now running on it.
Grok 4.6 is now available across Grok.com, iOS, and Android for everyone. It previously ran only in coding tools and Grok Build, but now every mode uses the new model.
Users are highlighting its strength in long-running tasks, research, and handling large data in regular conversations. SuperGrok subscribers are already testing Heavy mode for tougher problems, with Grok 4.7 rumored next.
Key Capabilities:

• All Modes Upgraded: Fast, Expert, Heavy, Auto, and Build now all run on Grok 4.6.
• Better at Long Tasks: Users report stronger performance on research and multi-step agentic queries.
• Cross Platform: Rolled out simultaneously on Grok.com, iOS, and Android for all users.

 Standard seats are free, and premium seats with 5x usage come at an 80% discount.
Starting today, 10,000 scientists across fields from math to chemistry to physics can get Claude through a new Claude Team plan. Standard seats are free, while premium seats with 5x usage limits cost $15 a month for one year.
Claude has shown recent progress on advanced physics calculations and protein design. Principal investigators can sign up and add researchers from their group, with plans to expand well beyond the initial 10,000 seats.
Key Capabilities:

• Free Access: Standard seats cost nothing for qualified researchers at academic and nonprofit institutions.
• Discounted Premium: Premium seats with 5x usage limits are $15 a month, an 80% discount.
• Built for Research: Targets scientific work like protein design and advanced physics calculations.

 GLM-5.3 weights are fully open, letting anyone run and customize the model themselves.
Z.ai released GLM-5.3 as fully open-weight, calling it their most capable model yet for agentic coding and cyber defense. Anyone can now download, run, and customize it freely.
It scores 50% better on internal code benchmarks and hits 84.5% on CyberGym for finding vulnerabilities, beating some closed models. Quantized versions run on consumer setups like a 256GB Mac Studio.
Key Capabilities:

• Coding Strength: Scores 50% better than prior versions on Z.ai's internal code benchmarks.
• Cyber Defense: Hits 84.5% on CyberGym, a benchmark for discovering security vulnerabilities.
• Runs Locally: Quantized versions from Unsloth and Ollama work on consumer hardware like a Mac Studio.

Cursor's direct access to OpenAI models ends November 12, following its acquisition by SpaceX.
OpenAI is ending its partnership with Cursor, the coding tool now owned by SpaceX. Under the proposal, Cursor's direct model access shuts down on November 12, giving developers time to switch.
The decision reportedly stems from trust concerns tied to SpaceX's history with compliance. Cursor says OpenAI models make up only 5% of its traffic, softening the blow.
Key Capabilities:

• Hard Cutoff: Direct OpenAI model access for Cursor ends completely on November 12, 2026.
• Small Impact: OpenAI models reportedly account for just 5% of Cursor's total traffic today.
• Workarounds Exist: Developers can switch to personal API keys or other AI providers to keep working.