Ox Alpha Revealed, OpenAI's Latest Pricing Updates, and Our Coding Model Tier List
Summary
Theo and Ben discuss OpenAI’s price cuts against Anthropic, arguing that GPT-5.6 Sonnet’s efficiency beats competitors like Kimi K3 despite higher token costs. They analyze the mysterious "Ox Alpha" model, revealing it as GLM-5.3 Flash, and debate the growing gap between raw intelligence and usability. The episode concludes with a head-to-head model tier list, where the hosts rank current AI models based on performance, speed, and value.
Topics discussed
Intro: Drinking game and episode setup
Sponsor: General Translation for product localization
GPT-4o-mini price cuts and speed comparisons
Anthropic pricing strategy and Opus vs Sonnet value
Rumors of unreleased Anthropic models and internal snapshots
OpenAI's competitive pressure on Anthropic and open weights
Future of 'potato mode' models and agent orchestration
Aux Alpha: New stealth model on OpenRouter
RL pipelines, DeepSeek V4, and model behavior tuning
Tier List: Muse Spark and contributor tier pricing
Tier List: Llama 3.2, Sonnet 5, and Opus 5
Tier List: Gemini 1.5 Flash and Pro models
Tier List: DeepSeek V4 Flash and open weight models
Tier List: Grok 4, Kimi K3, and local inference hardware
Tier List: GPT-4o-mini vs Claude Sonnet 3.5
Tier List: Claude Opus 4 and final rankings
Final Tier List reveal and closing thoughts
Listen ad-free on Castria