DeepSeek’s Oct 2025 open-source OCR model compresses text up to 10x via images, with reported efficiency gains of 7–20×, enabling million-token contexts.
Microsoft is expanding Copilot Voice, Vision, and Actions to all Windows 11 PCs, timed with Windows 10 end of support, ESU options through 2026, and renewed debate on security and e-waste.
Google launched Veo 3.1 on Oct 15, 2025, adding image-to-video with simultaneous audio, stronger prompt adherence, and granular editing for Flow, Gemini, and Vertex.
Anthropic introduces Claude Haiku 4.5 (default for free plans) combining low cost, high speed, extended thinking, and benchmark parity with top models, plus clear pricing and endpoint options for enterprise deployment.
Microsoft AI introduced MAI-Image-1, its first in-house image generator, ranked No. 9 on LMArena and focused on speed, photorealism, and reduced “AI-style” clichés ahead of Copilot and Bing integration.
Google launched Gemini Enterprise, a workplace AI hub that integrates with Workspace, Microsoft 365, Salesforce, and SAP, combining pre-built and custom agents with multimodal tools.
OpenAI’s Sora 2 debuts with advanced realism, audio sync, cameo controls, provenance signals, and a creation-first social feed, with API and Android on the way.
Anthropic released Claude Sonnet 4.5 (keeping Sonnet pricing) while adding Agent SDK, VS Code support, checkpoints, code execution, and Chrome automation to push AI into production coding.
Alibaba launches Qwen3-Max with trillion-parameter scale and adds open-source Qwen3-Omni, a real-time multimodal model with competitive benchmarks and clear API pricing.
