OpenAI Debuts GPT-5.5 as Claude AI Integrates Directly Into Microsoft Office
Frontier Model Developments
OpenAI Releases GPT-5.5 and GPT-5.5-Cyber: OpenAI has officially introduced the GPT-5.5 model family, including a specialized version designed for advanced cybersecurity tasks and trusted access. Fields Medal-winning mathematician Timothy Gowers reported using the Pro version to solve open problems, warning that AI's rapid progress may soon cause a "crisis" in traditional mathematical research.
- Fields Medal winning mathematician Timothy Gowers used GPT5.5 Pro to solve open problems, believes mathematical research will face a ‘crisis’ very soon with current rate of progress
- Scaling Trusted Access for Cyber with GPT-5.5 and GPT-5.5-Cyber
METR Evaluates Claude Mythos Preview: The METR evaluation of an early version of Claude Mythos suggests the model can maintain a 50%-time-horizon of at least 16 hours on complex tasks. This performance places the model at the upper limit of current measurable AI capabilities, though researchers note that task coverage at this range remains limited.
Product Integrations & Productivity
Claude AI Integrated into Microsoft Office Suite: Anthropic's Claude AI is now integrated directly into Word, Excel, PowerPoint, and Outlook. This update allows the assistant to maintain conversation context seamlessly as users switch between different Microsoft applications, enhancing workflow continuity.
Codex Coding Assistant Expands to Chrome and Mobile: OpenAI’s Codex now functions directly within the Chrome browser on macOS and Windows, allowing for parallel coding support across different tabs and sites. Additionally, a forthcoming update to the ChatGPT Android app will allow developers to remotely control Codex coding sessions on their PCs via their smartphones.
- The ChatGPT Android app should soon allow users to remotely control Codex coding sessions on their PCs
- Codex now works directly in Chrome on macOS and Windows.
Open-Source & Local AI
Release of Qwen3.6 35B Uncensored Variants: A new "uncensored heretic" version of Qwen3.6 35B has been released, featuring extremely low refusal rates and high retention of native MTPs. The model is available in several formats, including Safetensors, GGUF, and NVFP4, catering to various local LLM hardware setups.
- Qwen3.6 35B A3B uncensored heretic Native MTP Preserved is Out Now With KLD 0.0015, 10/100 Refusals and the Full 19 MTPs Preserved and Retained, Available in Safetensors, GGUFs. NVFP4, NVFP4 GGUFs and GPTQ-Int4 Formats
- https://huggingface.co/llmfan46/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved
- https://huggingface.co/llmfan46/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-GGUF
- https://huggingface.co/llmfan46/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-NVFP4-Experts-Only
- https://huggingface.co/llmfan46/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-NVFP4-Experts-Only-GGUF
- https://huggingface.co/llmfan46/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-GPTQ-Int4
- https://huggingface.co/llmfan46/models
Lemonade Platform Adds vLLM ROCm Support: Lemonade has introduced vLLM ROCm as an experimental backend, allowing users to run Safetensors-based LLMs directly on the platform. This addition provides a new engine for model exploration before committing to GGUF conversions.
Industry & Business
University of Michigan's Early OpenAI Stake: Reports indicate that the University of Michigan made a strategic $20 million investment in OpenAI long before its current valuation surge. This early stake, placed prior to the release of ChatGPT and Microsoft’s major involvement, is now estimated to be worth billions.