Thinking Machines Lab – Jan Sverre Bauge

Inkling 2 Small shown as a 594B Mixture-of-Experts model with 17B active parameters per token

Inkling 2 Small is enormous – but activates only 17B parameters per token

Inkling 2 Small has 594B parameters but activates only 17B per token. It is available on Tinker now, while the promised Apache 2.0 weights have not yet been released.

6. November 2026
4 views
7 minute read

Jan Sverre Bauge

I have worked with technology since childhood, from Amstrad computers and bulletin board systems to today’s AI tools. I build my own AI systems, automate content production, and keep experimenting with new tools and ways of working. My goal is simple: to make advanced technology understandable and accessible to everyone.

Gemini Omni 1.2 Flash generating a 15-second AI video through a conversational editing workflow

AI Models

Gemini Omni 1.2 Flash delivers 15-second AI video for $1.05

Gemini Omni 1.2 Flash creates and edits up to 15 seconds of AI video at $0.07 per…

Jan Sverre Bauge
6. November 2026
4 views
7 minute read

GPT-6 Cyber analyzing authorized targets while strict technical boundaries limit its reach

AI Models

GPT-6 Cyber finds more vulnerabilities – and stays in bounds more often

GPT-6 Cyber is OpenAI's most capable cybersecurity model. It found 81 serious Firefox…

Jan Sverre Bauge
6. November 2026
23 views
7 minute read

Google Search presented to schools as a choice between built-in AI features and no Search access

AI News

Google gives schools a fake choice: AI in Search or no Search

Google lets schools disable Gemini, but not the AI inside Search. Rejecting AI means…

Jan Sverre Bauge
5. November 2026
19 views
7 minute read

Kimi K3.1 Pro released with open weights for independent deployment and inspection

AI Models

Kimi K3.1 Pro now has open weights – Moonshot kept its promise

Kimi K3.1 Pro now has open weights. See what Moonshot released, how large the model is,…

Jan Sverre Bauge
5. November 2026
26 views
7 minute read