Written edition
Meta's $10M model outperforms Grok, resets AI coding wars
The last 72 hours show a frontier recalibrating across three axes: regulatory pressure tightening around open models and autonomy, capital-intensive bets on new hardware platforms facing cheaper alternatives, and most acutely, a shock to the frontier-model oligopoly from an unexpected source.
Meta's Llama Flash: $10M model outperforms frontier competitors
Meta released a sub-$10 million training run that beats Grok 4.5, upending the assumption that frontier AI coding models require mega-scale spending. The move democratizes access: where competitors needed vast resources to build competitive tools, any well-funded team can now train a peer model. This shatters the narrative of an inevitable oligopoly and forces every coding-AI vendor to justify premium pricing. For enterprises and smaller tooling companies, the implication is immediate—cheaper, open-weight alternatives are no longer theoretical.
