The new Inkling 2 Small model from Thinking Machines aims to challenge existing leaders in open-weight AI.
In an exciting development for the artificial intelligence landscape, Thinking Machines Lab has unveiled its newest model, Inkling 2 Small. This innovative creation is led by Mira Murati, a former CTO at OpenAI, and is positioned to shake up the dominance of current models primarily produced by Chinese companies.
Specifications and Capabilities
Inkling 2 Small boasts an impressive 594 billion parameters, enabling it to handle both image inputs and text outputs. While the model is praised for its capabilities in web and game design, it has also made notable strides in addressing coding challenges. According to the team, the model performs well in real agent loops, making it user-friendly in practical applications.
Performance Metrics
The model has been evaluated by Artificial Analysis, achieving a score of 31 on the Artificial Analysis Intelligence Index. This score indicates a competitive presence among similar models, tying with the Gemini 3.8 Flash. For context, leading models such as Gemini 4 Carbon and Claude Opus 5.5 lead the index with scores of 60 and 59, respectively.
The Growing Open-Weight AI Landscape
Thinking Machines plans to make the model's parameters available for public use next week, following a trend in the industry to ensure thorough safety assessments before releasing AI weights. This cautious approach is critical, especially given the rising concerns related to AI security. This follows the recent launch of French lab Mistral's Le Chonk model, which has quickly emerged as a competitor for coding tasks in the Western market.
Strengths and Challenges
While the Inkling 2 Small is making strides, it comes with its own set of challenges. Artificial Analysis reports a 41% boost in automation capabilities and a 20% improvement in coding proficiency compared to its predecessor. However, it has been noted that the model's performance is somewhat sluggish per task, and its tendency toward verbose responses could deter some users.
At a cost of $0.30 for every one million input tokens, the pricing is slightly above the median of $0.24 for similar models, potentially raising questions for budget-conscious users.
Improved Accuracy
Recent testing also revealed advancements in accuracy metrics. The rate of hallucinations has decreased significantly, from 61% down to 27%, and overall accuracy improved from 32% to 34%. Additionally, the attempt rate saw a reduction from 77% to 53%, demonstrating better alignment with expected outputs.
In conclusion, the launch of Inkling 2 Small from Thinking Machines marks a noteworthy entry into the Western open-weight AI market. With growing attention on the capabilities of AI technologies, this model's arrival is likely to influence the competitive dynamics, challenging existing leaders and encouraging further innovations.