The TokenSpeed engine now counts the cumulative number of reasoning tokens among committed tokens and exports it as `reasoning_tokens` to the output batch and clients, added by the commit of November 8, 2026.
The TokenSpeed engine now counts the cumulative number of reasoning tokens among committed tokens and exports it as `reasoning_tokens` to the output batch and clients, added by the commit of November 8, 2026.