MondAI Roundup - September 2026

September's MondAI Roundup covers two months of developments. There were enough model releases that some had already been superseded by the time of the session. I look at Claude Fable 5.1 and GPT-6 Astra, improvements to free chat services, and the difficulty of comparing models using benchmark scores alone.

The open-model section covers what can run on personal hardware and what needs a much larger computing budget. The practical examples include transcription, voice models that delegate work to other models, and agents that carry out scheduled tasks or communicate across sessions. There is also a shorter look at image generation and world models.

The industry section covers NVIDIA's agreement to acquire Hugging Face, Stripe's agreement to acquire OpenRouter, and the consequences of Cursor's completed acquisition by SpaceX. The security chapter distinguishes the incidents reported by OpenAI, Anthropic and the UK AI Security Institute, and explains the discussion about independent evaluation. The final chapter covers the debate about AI in mathematics.

Key Stories

• Fable and Mythos 5.1 and Astra bring stronger science and computer-use results, with different safeguards and conditions of access.

• Muse Glimmer and Qwen3.8-27B expand the options for local coding and image analysis.

• GPT-Live-1 separates the work of handling a voice conversation from deeper reasoning delegated to another model.

• The independent Hugging Face investigation examines agents acting beyond their evaluation tasks, including their communication and collective work.

Resources

• Detailed notes contain ten chapters based on the session transcript.

• Presentation slides contain the individual news items and source links.

• Quick reference index lists people, organisations and resources.

Key Links

• Artificial Analysis Intelligence Index v4.3 explains the updated model comparison.

• OpenClaw 2.0 describes the installation and shared-session changes.

• We Must Pace the Frontier sets out Dario Amodei's proposals, including embedded independent evaluators.

• A Severe Misalignment of AI in Mathematics discusses the relationship between solving problems and developing mathematical understanding.

Next Session

The next MondAI Roundup is on Tuesday 13 October 2026 at 12:30 BST, exceptionally on a Tuesday rather than the usual Monday. Register for the October session.