Thinking Machines Unveils Inkling 2 Small, Top Open AI Model
Thinking Machines launches Inkling 2 Small, a groundbreaking open AI model that challenges competitors from China and excels in coding.
Introduction of Inkling 2 Small
Thinking Machines Lab, led by former OpenAI CTO Mira Murati, has launched its highly anticipated AI model, Inkling 2 Small. This innovative model is positioned to compete aggressively with current leaders in open AI models, particularly those from China. According to evaluations from Artificial Analysis, this release could shift the competitive dynamics in the open model sector.
Key Features of the Model
Inkling 2 Small boasts an impressive 594 billion parameters and is designed to handle image inputs, although it only outputs text. The creators emphasize its capabilities in web and game design while highlighting its proficiency in addressing coding challenges effectively. During testing, the model demonstrated improvements in code generation, making it more user-friendly for integration into various applications.
Performance Metrics
On the Artificial Analysis Intelligence Index, Inkling 2 Small achieved a score of 31, indicating strong capabilities relative to its peers. This score positions it alongside Gemini 3.8 Flash and signifies a respectable performance in comparisons made across categories, including coding and scientific applications. For perspective, top models like Gemini 4 Carbon and Claude Opus 5.5 rate at 60 and 59 respectively.
The Competitive Landscape
Thinking Machines plans to release the weights for Inkling 2 Small next week, a strategic move that aligns with current practices in the AI field. Companies are increasingly opting for a phased release, conducting tests alongside cybersecurity teams before making their models available to the public.
The release follows the successful launch of the French lab Mistral's Le Chonk model, which has made notable advancements in coding performance, establishing a competitive benchmark for western-developed open models.
Market Implications
According to Artificial Analysis, the opportunity to make Inkling 2 Small's weights accessible could place it at the forefront of open models developed outside of China, based on current rankings. However, it is important to note that while this model shows significant strengths, it also has drawbacks.
Evaluating Strengths and Weaknesses
Despite its impressive capabilities, Inkling 2 Small faces criticism for being slower per task and somewhat verbose. Its pricing structure also raises eyebrows, with a cost of $0.30 per one million input tokens, which is above the median price of $0.24 for similar models.
Nevertheless, its improvements over its predecessor are commendable, with a 41% increase in automation efficiency and a 20% improvement in coding performance noted by reviewers. These advancements demonstrate significant progress within its model category.
Accuracy Enhancements
In terms of accuracy, the developments are noteworthy. The model's hallucination rate decreased from 61% to 27% during evaluations on the AA-Omniscience v1.1 test suite. Additionally, its accuracy showed a slight increase from 32% to 34%, while the attempt rate declined from 77% to 53%, signaling better alignment to expected outputs.
Conclusion
In conclusion, Inkling 2 Small from Thinking Machines Lab represents a significant advancement in the landscape of western open-weight AI models. While it still faces challenges in areas like speed and pricing, its capabilities in coding and intelligent processing highlight its potential to become a major player in the AI landscape.
