The latest DesignArena ranking places MiniMax H3-Image second among open-weight image models, averaging 15 seconds per generation, making it the fastest model in the top 3.
Anthropic has announced its initial public offering, having filed a registration statement with the SEC, with Goldman Sachs and Morgan Stanley involved in distributing the preliminary prospectus.
Anthropic has launched an office productivity suite for Claude, featuring spreadsheets, whiteboards, and data analysis with real-time collaboration and third-party integrations. Microsoft's stock fell following the announcement, amid concerns over potential disruption to its long-standing Office business.
ByteDance has launched Seedance 2.5 flash, a lightweight video generation model that doubles generation speed while maintaining quality close to the flagship version and significantly reducing costs. It supports multimodal inputs and longer video generation, and is now rolling out on platforms such as Jimeng and LibTV.
Alibaba has launched the preview of Qwen4-Max, debuting on the Qianwen AI Platform Token Plan. The model features comprehensive upgrades in world knowledge, perception, coding, and agentic capabilities, excelling in complex programming, professional office tasks, and design. Internal tests highlight strong performance in 3D modeling and human-computer interaction.
Seedance 2.5 Flash has been globally launched on LibTV, accessible to all users. The model focuses on faster generation speeds and lower costs, supporting video creation from 4 to 30 seconds in 480p and 720p resolutions. Features include text-to-video, start-and-end frame generation, multi-material reference, audio-enabled video production, as well as video editing and extension. Pricing aligns with the Seedance 2.0 Standard edition, with 720p rates dropping to 0.10 yuan per second during the Do
DeepMind has published a blog post introducing AlphaProof Code, described as a breakthrough in verifying complex software at depth; the same page also lists CodeMender, an AI agent for code security.
Google DeepMind's official blog has published an article titled Gemini Robotics 2.5 accelerates real-world general dexterity, alongside a listing for Gemini 4 Argon for Robotics, without disclosing technical details or test results.
@Re7_AI published an updated Grok Bot usage guide on 29 October 2026, saying the earlier tutorial was nearly half outdated. The new version is organised into three stages: day-one acceptance, running real tasks with ready-made templates in week one, and later quota management plus memory cleanup. The post is community advice; official first-run steps live in xAI documentation.
Replying on X to a question about Grok 4.8's timing, Musk said pre-training is finished, the RL and inference stack was rewritten from scratch in C/C++, and the model is now in RL with evaluations just beginning.
Google's Gemini 4 Carbon model API is now live, available to Ultra subscribers. The model ranks first in multiple benchmarks, offers significantly lower task costs than competitors, and features automatic 3D model reconstruction and multi-agent collaboration scheduling.
MiniMax has officially open-sourced its image generation model, MiniMax H3-Image, which supports text-to-image, image reference, and editing capabilities. The model can generate 4096x4096 high-resolution images within 10 seconds, operating at approximately five times the speed of mainstream market standards. It boasts the strongest instruction-following capability among open-source models and supports up to 16 reference images.