Thinking Machines' AI Model Challenges Chinese Dominance in Open-Weight AI

A new AI model has emerged as a top contender in the field of open-weight artificial intelligence, challenging the dominance of Chinese companies that have led the pack for several years. According to third-party tester Artificial Analysis, Thinking Machines' latest model, Inkling 2 Small, is poised to give its Chinese counterparts a run for their lead.

Chinese companies such as DeepSeek and Moonshot have been at the forefront of open models in recent years, consistently ranking high on leaderboards and rivaling the capabilities of more expensive closed models from OpenAI and Anthropic. However, with DeepSeek's R1 model performing similarly to OpenAI's o1 in January 2025, Chinese open models have started to give investors and enterprise customers reason to reconsider their contracts with US-based labs.

The emergence of Western companies creating competitive open models is significantly altering the landscape. Thinking Machines Lab, a small AI startup founded by ex-OpenAI CTO Mira Murati, has released its latest model, which boasts 594 billion parameters and supports image inputs but outputs only text.

Thinking Machines has highlighted its web and game design capabilities, as well as its approach to coding problems, stating that the model is both stronger at coding and easier to use inside real agent loops. The company noted that it will release its weights next week, making its model publicly available for evaluation and testing.

The performance of Thinking Machines' new model, Inkling 2 Small, has been impressive in western open-weight AI, scoring a 31 on the Artificial Analysis Intelligence Index. This composite test evaluates progress toward artificial general intelligence by combining benchmark scores from various categories such as coding, science, and general knowledge. The score of 31 is notable, exceeding average performance for models in its class and matching that of Gemini 3.8 Flash.

The company's decision to release the model's weights next week follows a common practice in the industry, where companies test their open models with cybersecurity teams and safety reviewers before making them publicly available. This approach has become increasingly prevalent due to growing concerns about AI security. The staggered release process allows for a more thorough evaluation of the model's potential vulnerabilities.

In related news, Mistral's Le Chonk model has made significant strides in coding abilities, now rivaling OpenAI's GPT-6.1 Terra. This achievement is particularly noteworthy given that it is rare for western-made open models to reach such levels of performance. The comparison highlights the progress being made by AI labs outside China.

The impending release of Thinking Machines' weights is expected to further solidify its position as a leading developer of open-weight AI, according to current rankings. This development would mark a significant milestone in the field, with Artificial Analysis noting that it would be the highest-scoring open weights model developed outside China.

The latest AI model from Thinking Machines, Inkling 2 Small, has taken a significant lead in western open-weight AI. According to reviewers at Artificial Analysis, this model stands out for its impressive performance and relatively affordable pricing, despite being somewhat expensive compared to its peers.

Inkling 2 Small has made notable strides over its predecessor, with a 41% increase in automation capabilities and a 20% boost in coding abilities. However, it's worth noting that many AI models excel in specific areas or on particular benchmarks, making direct comparisons challenging.

The model's performance gains are also evident in its accuracy improvements, as seen in the AA-Omniscience v1.1 benchmark. Here, Inkling 2 Small has reduced hallucination rates and improved overall accuracy compared to its predecessor.

Overall, this development marks a significant milestone in AI research, with Artificial Analysis noting that it would be the highest-scoring open weights model developed outside China.