Thinking Machines Unveils Inkling 2 Small, Raises AI Stakes

Thinking Machines Lab, founded by ex-OpenAI CTO Mira Murati, has introduced its latest AI model, Inkling 2 Small. This new model is poised to alter the competitive landscape, challenging the dominance of Chinese offerings in the open-weight AI segment.

What Makes Inkling 2 Small Stand Out?

Equipped with an impressive 594 billion parameters, Inkling 2 Small not only supports image inputs but is also tailored to produce text outputs. The model stands out for its capabilities in web and game design, as well as its adeptness at tackling coding challenges. According to its developers, its functionality within real-time agent loops enhances user experience, making it more accessible for developers.

Performance Insights

Recent evaluations by Artificial Analysis rated Inkling 2 Small with a score of 31 on the Artificial Analysis Intelligence Index, a benchmark assessing traits related to artificial general intelligence. This score places it among high-performing models, alongside Gemini 3.8 Flash. To provide context, the leading models, Gemini 4 Carbon and Claude Opus 5.5, scored 60 and 59, respectively.

As for its release strategy, Thinking Machines plans to publish the model's weights soon, aligning with a growing trend in the industry that emphasizes thorough safety checks prior to the public unveiling of AI systems. This cautious approach corresponds with increasing global concerns regarding AI security and reliability.

Catching Up with the AI Landscape

Historically, Chinese companies have spearheaded developments in open-weight models, capturing significant market share. Notably, both DeepSeek and Moonshot frequently outperform many traditional models, proving their prowess against pricier, closed-off alternatives from companies like OpenAI and Anthropic.

In a similar vein, French lab Mistral recently released its Le Chonk model, which has quickly positioned itself as a formidable competitor, especially in coding proficiency—a rare achievement for Western models.

Advantages and Areas for Improvement

While the performance of Inkling 2 Small has shown notable enhancements over its predecessor—reporting a 41% increase in automation and a 20% jump in coding capabilities—there are still challenges to address. Critics from Artificial Analysis cited its performance as being relatively sluggish for tasks and mentioned a tendency towards verbosity, which could impact user satisfaction.

Pricing for Inkling 2 Small stands at $0.30 per million input tokens, slightly above the median of $0.24 within its category. This could pose budgetary considerations for some prospective users.

Improved Accuracy Metrics

The results also indicated progress in accuracy, with reductions in the hallucination rate falling from 61% to 27%. Correspondingly, accuracy increased slightly from 32% to 34%, while the attempt rate decreased from 77% to 53%. Such improvements indicate a better alignment between output and expected results, showcasing the model's evolving reliability.

Inkling 2 Small's emergence on the open-weight AI stage represents a significant step for Thinking Machines and the wider Western AI community. As the competition heats up, this model could influence the development of future AI technologies, encouraging innovation and enhanced capabilities.