Third-Party Evaluation Released: Thinking Machines' New Model Ties with GPT-Realtime-2, Claims Top Spot on Audio Leaderboard

According to Dynamix Beating Monitor, data platform Scale Labs today released the latest Audio MC S2S rankings. The evaluation results show that the newly released TML-Interaction-Small model by Thinking Machines achieved an APR score of 43.4%, tying for first place with OpenAI's GPT-Realtime-2 (xHigh).

In terms of specific scores, GPT-Realtime-2 (xHigh) led the absolute score with 48.45 points, while TML-Interaction-Small followed closely behind with 43.36 points. As the score difference between the two falls within the margin of error, they were officially deemed tied for first place. The second tier that followed consisted of the standard version of GPT-Realtime-2 (37.61 points), the Gemini 3.1 Flash Live in think mode (36.06 points), and the previous version GPT-Realtime-1.5.

Scale Labs commented that the model, while maintaining conversation response speed, demonstrated a rare long-context awareness ability among existing bidirectional models.