Will any Thinking Machines Lab model rank in the top 10 of the Artificial Analysis Intelligence Index at any point before January 1, 2028?

probability: 13%

rationale:

Thinking Machines Lab (TML) faces a steep climb to crack the top 10. Its current model, Inkling (released July 15, 2026), scores roughly 40.7–41 on the Artificial Analysis Intelligence Index v4.1, placing it well outside the top tier 44 sources. The current threshold for the top 10 is around 56, leaving TML with a ~15-point deficit today. Furthermore, the index ranks individual reasoning-effort configurations separately, meaning the top 10 is crowded by multiple variants from a small handful of frontier leaders like Anthropic and OpenAI artificialanalysis.ai. TML would not just need to beat other open-weights models; it would need to reach near-frontier quality overall against a bar that is continuously rising.

While TML has secured massive compute resources, the timelines for data center deployments and hardware deliveries create a tight window for releasing a fully trained model before 2028. The 1 GW Nvidia Vera Rubin deployment is targeted to begin in early 2027 thinkingmachines.ai. Given the time required to ramp up capacity, execute a multi-month pretraining run, and conduct extensive reinforcement learning (bolstered by its Google Cloud AI Hypercomputer expansion 2 sources), a flagship model trained on this new hardware is unlikely to be evaluated and released before late 2027 or early 2028. Any model released in mid-2027 would likely be trained on pre-Rubin capacity, requiring an unprecedented relative capability jump to reach a top-10 threshold that will likely have moved into the 60s.

TML's organizational posture also limits its benchmark-chasing potential. The lab has explicitly stated that Inkling is not intended to be the strongest overall model, positioning it instead as a broad, customizable substrate monetized through its Tinker fine-tuning product 66 sources. Additionally, TML has experienced notable talent attrition, including the departure of co-founders and key researchers to competitors en.wikipedia.org, which may negatively impact research velocity. An external analysis similarly projects TML's first top-5 frontier model will not arrive until a median of December 2030 futuresearch.ai, consistent with a top-10 entry occurring mostly after 2027.

Despite these structural and temporal obstacles, there are plausible paths to a top-10 appearance. A compute-fueled, RL-heavy release could close the gap faster than expected, as other open-weight labs like Moonshot, Qwen, and DeepSeek have demonstrated artificialanalysis.ai. The resolution criteria only require a model to appear in the top 10 at any single moment, meaning a launch-day spike could briefly secure a spot before incumbents respond. Furthermore, index rebasing or deduplication by model family could mechanically lower the top-10 threshold, offering a modest boost to TML's chances. However, given the massive current deficit and the tight 17-month window dictated by timelines for data center deployments and hardware deliveries, a fully trained model is unlikely to be ready in time to capitalize on these upside scenarios, anchoring the probability at 13%.