Thinking Machines Launches Cutting-Edge Inkling 2 Small AI Model

Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, has unveiled its latest project, the Inkling 2 Small. This innovative AI model is set to shift the competitive landscape in the realm of open AI models, particularly as it aims to contend with leading products from Chinese companies.

Overview of Inkling 2 Small

With an astounding 594 billion parameters, Inkling 2 Small exhibits remarkable capabilities, particularly in coding and design tasks. Although it can handle image inputs, its output is strictly text-based. The team emphasizes its proficiency in web development and gaming, making it an adaptable tool for various applications.

Performance Assessment

Artificial Analysis, an independent reviewer, awarded Inkling 2 Small a score of 31 on the Artificial Analysis Intelligence Index. This score indicates its competitive edge among similar models, aligning it with the performance of Gemini 3.8 Flash. For reference, top-ranking models like Gemini 4 Carbon and Claude Opus 5.5 currently boast scores of 60 and 59, respectively.

The Competitive Context

Thinking Machines plans to release the model’s weights next week, a standard practice in the AI industry today. This approach allows companies to rigorously test their models with cybersecurity and safety experts before making them publicly available. The announcement follows the recent launch of Mistral's Le Chonk model, which has raised the bar for coding capabilities among models developed outside of China.

Market Position and Challenges

By preparing to release Inkling 2 Small’s model weights, Thinking Machines could find itself leading the pack in open-weight models created in the West, according to Artificial Analysis. However, challenges remain. Despite its strengths, reviewers noted that Inkling 2 Small tends to be on the slower side and can produce verbose outputs. Its pricing structure, set at $0.30 per one million input tokens, is above the category average of $0.24, which may deter potential users.

Improved Functionality

Improvements over its predecessor are notable. Inkling 2 Small has achieved a 41% boost in automation efficiency and a 20% enhancement in coding skills. This advancement signifies its evolving capabilities and potential for real-world applications.

Accuracy Developments

In terms of accuracy, the model has also made strides. Testing results showed a decrease in the hallucination rate from 61% to 27% across the AA-Omniscience v1.1 benchmarks. Moreover, its accuracy improved slightly from 32% to 34%, while the attempt rate dropped from 77% to 53%, indicating a smoother alignment with expected outputs.

Final Thoughts

In conclusion, Inkling 2 Small signals a significant evolution in the open-weight AI landscape produced in the West. While aspects like speed and pricing need reconsideration, its capabilities in coding and intelligent processing highlight its potential to establish itself as a key player in the AI domain.