Interconnects Open models: Hot or Not with Nathan Lambert & Florian Brand
62 snips
Dec 18, 2025 Florian Brand, an open model analyst, joins Nathan Lambert for a fiery debate on the evolving landscape of open models. They dive into which labs have gained or lost momentum, tackling the fall of Llama and Huawei's controversial model repackaging. The duo explores the significance of big tech involvement, evaluates chaotic open releases, and predicts the future of GPT-OSS. They also discuss the balance between American and Chinese models, highlighting the rise of underdogs and the impact of coding performance on adoption.
AI Snips
Chapters
Transcript
Episode notes
GPT-OSS Is Powerful But Fragile
- GPT-OSS is powerful yet brittle: optimal only when used exactly as intended.
- Florian hopes for a GPT-OSS 2 that is more robust across setups and languages.
Grow Users With Free, Useful Releases
- Release useful models and make them freely accessible to build user base.
- Florian suggests the 'Chinese model playbook' of free releases to rapidly gain users.
Minimax Dogfoods Its Coding Models
- Minimax dogfoods its own models and uses them as primary coding models.
- Nathan and Florian note strong adoption of Minimax M variants for coding tasks.