Matthew Berman and Theo Jaffee discuss GPT-5.6 Sol, Terra and Luna as a model family rather than treating a single benchmark ranking as the whole story. Matthew Berman describes reliable browser use and multi-day coding work, while both speakers emphasize the distinction between frontier planning and cheaper execution.
Matthew Berman and Theo Jaffee compare the workhorse-model strategy associated with Meta Muse Spark and Grok against the ambition to build the strongest frontier model. Their disagreement centers on whether frontier capability can also produce better small models through distillation and how enterprises will respond to growing token bills.
Matthew Berman and Theo Jaffee discuss the combination of coding data, application usage and compute capacity in xAI's strategy. They treat future training runs and model performance as uncertain, and argue that continuing competition matters for pricing and market power.
Watch on YouTube
