Meta (META) says New Spark has become a top-two model on Open Code go, powering over 120mln sessions and 79tln developer tokens

Self-reported adoption metrics from a model developer are a recurring feature of the current AI capex cycle, and the historical read is that headline session and token counts function more as positioning than as independently verifiable traction: the figures come from the issuer, the definitions are chosen by the issuer, and comparable claims from rival labs have tended to arrive in quick succession around launches. The claims worth separating are usage volume versus model quality, since high token throughput on free or open-weight offerings has historically coexisted with weak monetisation, and a top-two ranking on a single third-party platform says little about share across the broader developer market. For Meta specifically, the established pattern is that open-source model releases serve the wider strategy of commoditising the layer competitors charge for, so the market-relevant question is whether engagement of this kind feeds the core advertising and recommendation stack rather than whether it leads a leaderboard. The follow-ons that have mattered in past episodes are independent benchmark replication, enterprise uptake disclosures, and any commentary on inference cost per token, which is where usage at this scale becomes a margin question. As a single-company product announcement with unaudited figures, the signal is directional at best.