Murati's Thinking Machines releases 975B-parameter Inkling model

The more telling detail is the deliberate framing as a fine-tuning foundation rather than a frontier model. Murati is not competing on raw leaderboard position; she is betting that the next margin pool sits in customizable infrastructure for enterprise developers, not in chasing Chinese labs on aggregate benchmarks.

This is largely disconnected from the recent coverage on this site, which has focused on user-facing consent and default settings (see the WIRED piece on opting out of AI features from July 16). That debate is about how AI reaches end users; Inkling operates one layer below, targeting developers who build those products. The relevant competitive context sits outside our current archive: the ongoing benchmark gap between U.S. and Chinese frontier labs, and the broader market question of whether open-weights releases at this scale can sustain a commercial business. Thinking Machines is essentially making a structural bet that developer lock-in through fine-tuning workflows is more durable than raw performance leadership.

Watch whether enterprise teams publicly report fine-tuned Inkling variants outperforming closed models on domain-specific evals within the next two quarters. If that evidence accumulates, the infrastructure-over-capability thesis holds; if adoption stays thin, the benchmark gap with Chinese models will look less like a strategic choice and more like a ceiling.