AI News from October 6, 2026

OpenAI shares mathematics results from an internal frontier model, along with Lean formalizations and research details, giving builders more to inspect.

OpenAI and Ironclad are training and evaluating AI agents on complex contract workflows, testing a path toward more capable computer use at work.

Mistral AI has unveiled Mistral Large 4, a 1T-parameter multimodal model that targets open and closed rivals while delaying weight release for safety testing.

Mistral has launched Mistral Large 4 as an open-weight model, giving AI teams another model to assess for controlled deployment and customization.

Reports say DeepSeek is weighing a funding round of up to $15 billion, signaling rising investor appetite for Chinese AI despite disclosure gaps.

Mistral’s CEO says a new AI model outperforms Chinese rivals in areas including cybersecurity, but independent benchmark details remain undisclosed.

GitHub’s open ReviewBench benchmark targets a clearer way to test AI code review agents, giving builders and buyers a common evaluation starting point.

Reka AI’s Rho-1 combines text, images, video, and robot control in one 19B research model, signaling a push toward unified AI systems.

DeepSeek is reportedly seeking at least $12 billion from Tencent, CATL and other backers, a potential funding test for China’s AI infrastructure race.

OpenAI is reported to be adding invisible watermarks to ChatGPT text in the EU, a compliance move that could reshape AI-content disclosure workflows.

A Gizmodo headline flags a new threat to OpenAI and Anthropic, but the supplied record omits the article’s evidence and does not identify it.

Reflection AI has released Beam, its first open model, sharpening the U.S.-China contest over accessible AI systems and expanding choice for builders.