OpenAI AI Updates: October 7, 2026

1. OpenAI Publishes 722 Math Manuscripts From an Internal Model, With Partial Lean Formalization

OpenAI. OpenAI released 722 manuscripts, grouped into 372 problem families, produced by an internal frontier model that was posed roughly 4,000 open problems across number theory, complexity theory, geometry, and mathematical physics. The openai/math GitHub repository (Apache-2.0) includes Lean proof formalizations and abridged reasoning summaries, with each accepted result using compute equivalent to about three hours of ChatGPT Pro thinking on average. The repository itself notes that not every result has a Lean formalization and that some unformalized results “could have issues,” so only the Lean-checked proofs can be verified mechanically without human refereeing. Source

2. OpenAI and Ironclad Train Computer-Use Agents on Contracting Workflows

OpenAI. OpenAI described how it is training and evaluating AI agents on complex contracting workflows inside Ironclad’s contract management software, as part of its push to improve computer use for professional work. The approach uses a real enterprise application, rather than generic browser benchmarks, as both the training environment and the evaluation target. For teams building agents on top of SaaS tools, it signals that OpenAI’s computer-use gains may increasingly be tuned to specific vendor applications. Source

3. Atlassian and OpenAI Connect ChatGPT and Codex to Jira and Confluence via MCP

OpenAI. Atlassian and OpenAI expanded their partnership so that GPT-6 family models power agents across Atlassian’s platform and its Rovo assistant, combined with Atlassian’s Teamwork Graph context layer spanning Jira, Confluence, and Bitbucket. The integrations include the Atlassian MCP Server in both ChatGPT and Codex, deeplinks from Atlassian to Codex as a coding tool, and OpenAI model selection in Rovo Agents; more than 3,000 Atlassian developers already use Codex. The companies are also exploring Jira integrations for assigning work to AI agents, tracking progress, and reviewing results. Source