Thinking Machines' Inkling 2 Small Leads Western Open-Weight AI Models

Thinking Machines has unveiled a new AI model named Inkling 2 Small, which has quickly risen to prominence in the western open-weight AI landscape. This model boasts an impressive 594 billion parameters, marking a significant leap from its predecessor ¹²³. According to reports, Inkling 2 Small has also achieved a notable score of 31 on the Artificial Analysis Intelligence Index, solidifying its position as a leader in the field ¹²³.

Inkling 2 Small reportedly excels especially in agentic tasks, showcasing advanced capabilities in decision-making and problem-solving ¹. Additionally, the model has demonstrated a 41% improvement in automation capabilities and a 20% boost in coding abilities compared to its previous version ². These enhancements have made Inkling 2 Small a standout choice for developers and researchers seeking powerful AI solutions.

One of the key features of Inkling 2 Small is its availability for fine-tuning and inference on the Tinker API, allowing users to customize and optimize the model for specific applications ¹³. Furthermore, the model's pricing structure is designed to be accessible, with costs set at $0.30 per 1 million input tokens and $1.00 per 1 million output tokens ³. This makes it a viable option for both individual users and large-scale enterprises.

However, Inkling 2 Small's performance is not without its challenges. While it excelled in certain areas, such as achieving 86.3% accuracy on SWE-Bench Pro v2 (Hard), it faced some difficulties in others. Specifically, the model performed poorly on Terminal-Bench 4.0, with an accuracy rate of just 13.1% ³. Despite this, the internal misbehavior score of Inkling 2 Small has significantly decreased from 66.8 to 14.9, indicating a more stable and reliable operation ³.

What We Know

• - Thinking Machines released Inkling 2 Small, a new AI model with 594 billion parameters ¹²³
• - Inkling 2 Small scored 31 on the Artificial Analysis Intelligence Index ¹²³
• - It excels particularly in agentic tasks ¹
• - The model improved by 41% in automation capabilities and 20% in coding abilities compared to its predecessor ²
• - Inkling 2 Small is available for fine-tuning and inference on Tinker API ¹³
• - Pricing is set at $0.30 per 1 million input tokens and $1.00 per 1 million output tokens ³
• - Internal misbehavior score decreased from 66.8 to 14.9 ³

What Remains Unclear

• - The exact number of active parameters in Inkling 2 Small ¹³
• - Whether Inkling 2 Small will continue to improve its performance in all benchmarks over time ³