China caught up to GPT-5.5 — and the model moat is gone
chinese models like kimi k3 and glm-5.3 now match top us models on benchmarks, shrinking the performance gap. investors worry that as raw model performance ceases to be a defensible moat, western labs must shift value to the systems surrounding their models.
chinese labs have released open-weights models like kimi k3 and glm-5.3 that rank near the top of broad benchmarks, matching proprietary us leaders like gpt-5.5 in areas like long knowledge tasks and agentic workflows. this rapid catch-up, exemplified by k3's debut in third place on artificial analysis's intelligence index, has narrowed the performance gap to the point where it is becoming a significant concern for investors and companies like anthropic ahead of their ipos. western labs allege that this parity is achieved through distillation and benchmaxxing, but regardless of the cause, the economic moat of raw model performance is disappearing. the industry is consequently pivoting its competitive focus from the model itself to the broader infrastructure and systems required to train and deploy them.