The one thing
OpenAI reports its internal model made progress on unsolved maths problems
Why it mattersOpenAI's unreleased model is producing checkable maths results, so expect that reasoning strength in products you can buy later.
Seventy: Wednesday, 7 October 2026
What happened. OpenAI released findings on unsolved maths problems worked by one of its internal frontier models. Machine-checked Lean versions of the proofs, plus supporting write-ups, are posted on GitHub.
Atlassian links OpenAI's top models to its workplace tools
Why it mattersAtlassian's tools are getting OpenAI models built in, so check what your team already pays for before buying separate AI add-ons.
What happened. Atlassian and OpenAI have widened their working relationship so OpenAI's most advanced AI can draw on a company's stored knowledge. The aim is to help teams organise, create and finish their work.
03Agents
OpenAI teams with Ironclad to build agents that handle contract work
Why it mattersOpenAI is using Ironclad's contract workflows as a proving ground for agents that operate software, so legal tooling is an early target.
What happened. OpenAI is working with contract software maker Ironclad to teach and test AI agents on multi-step contract tasks. The aim is to improve how agents operate software for professional work.
04Models
Amazon Bedrock adds Z.ai's 753B-parameter GLM 5.3 coding model
Why it mattersA 753B coding model now runs on , so you no longer need your own stack.
What happened. Z.ai's (Zhipu AI) GLM 5.3 is now available on Amazon Bedrock. It is built for coding and long-running agentic tasks, and AWS customers can use it without setting up their own servers for it.
NVIDIA survey: 89% of telecom firms call open models important to AI plans
Why it matters89% of telecom respondents in NVIDIA's own survey call open models important, so expect operator customers to ask for them.
What happened. NVIDIA's State of AI in Telecommunications report finds 89% of surveyed operators call open source models and software important to their AI strategy. NVIDIA says telcos want to trust, control and tailor AI for core uses such as self-running networks and customer support.
AWS lets SageMaker Studio launch JupyterLab and Code Editor on HyperPod clusters
Why it mattersYour data scientists can start work on HyperPod clusters in a few clicks, with no command-line setup or platform-team ticket.
What happened. Amazon SageMaker Studio now lets data scientists and ML engineers create and manage Spaces on HyperPod EKS clusters from the browser. They can launch JupyterLab and Code Editor environments without command-line tools.
07Agents
AWS shows how to build a voice travel assistant on Bedrock and Nova Sonic
Why it mattersAWS now documents a full reference build for voice agents that sit on top of your existing booking backend.
What happened. AWS published a tutorial for building a voice travel concierge on Amazon Bedrock AgentCore, a Managed Knowledge Base and Nova Sonic. It targets airline tasks such as changing a seat or checking a delay by speech.
That's the whole issue.
7 stories, about 2 min 27 sec. Nothing was left out to fill the time.
Words of the day
- Multi-agent system
- Several AI agents working on one task, each with its own job — one plans, one searches, one checks — passing work between them.
- Interpretability
- Work on understanding why a model produced a particular answer, rather than treating it as a black box.
- brand suitability controls
- Advertiser settings that restrict which kinds of content or contexts their ads can appear next to.
“Stay hungry. Stay foolish.”