Thinking Machines Inkling 2 Small: A Game-Changer for Western Open AI

Quick Verdict

Thinking Machines’ Inkling 2 Small is a significant breakthrough in the Western open-weight AI landscape. Developed by an OpenAI veteran, this 594-billion parameter model positions itself as a robust challenger to the dominance of Chinese open models, particularly in coding and automation. While it comes with a slightly higher price tag and a tendency for verbosity and slower task completion, its impressive leap in capabilities and commitment to open-weight release make it a compelling option for developers and businesses looking for powerful, accessible AI solutions outside of closed ecosystems.

Introduction: Challenging the Status Quo

For the past few years, the open-weight AI leaderboard has been largely dominated by models emerging from China, such as those from DeepSeek and Moonshot. These models have consistently offered capabilities that rival more expensive, closed-source offerings from Western giants like OpenAI and Anthropic, giving enterprises compelling reasons to reconsider their AI investments. Enter Thinking Machines Lab, a nimble startup spearheaded by former OpenAI CTO Mira Murati. Their latest creation, Inkling 2 Small, aims to disrupt this pattern and firmly plant a Western flag in the competitive open-weight arena.

Inkling 2 Small, a technically proprietary model until its weights are released in the coming week, has already garnered significant attention from third-party testers like Artificial Analysis. Their early assessments suggest it's poised to become the highest-scoring open-weight model developed outside China, marking a pivotal moment for the industry. This release, following French lab Mistral’s coding-focused Le Chonk, signals a growing resurgence of competitive open-weight AI development in the West, promising a more diverse and innovative future for AI accessibility.

Core Capabilities and Performance

Inkling 2 Small is an impressively scaled model, boasting 594 billion parameters. It offers multimodal input, specifically supporting image inputs, though its outputs are exclusively text-based. Thinking Machines highlights its particular strengths in web and game design, indicating a targeted utility for creative and development-focused applications. Crucially, the company emphasizes its approach to coding problems, stating that the model was evaluated not just on its generated code but also on its problem-solving methodology. This focus has resulted in a model that is "both stronger at coding and easier to use inside real agent loops," suggesting practical advantages for developers integrating AI into complex workflows.

On the Artificial Analysis Intelligence Index, a comprehensive test designed to track progress toward artificial general intelligence across coding, science, and general categories, Inkling 2 Small achieved a score of 31. This places it firmly above average for models in its class and notably ties it with Gemini 3.8 Flash. While this score doesn't reach the heights of top-tier models like Gemini 4 Carbon (60) or Claude Opus 5.5 (59), it represents a strong showing for an open-weight model, especially when considering its overall competitive positioning.

Compared to its predecessor, Inkling Small, the new model demonstrates significant progress. Artificial Analysis reviewers noted a remarkable 41% jump in automation capabilities and a 20% improvement in coding abilities. This iterative advancement underscores Thinking Machines' commitment to rapid innovation. Furthermore, the model has made strides in accuracy and reducing hallucinations. On the AA-Omniscience v1.1 test, Inkling 2 Small’s hallucination rate plummeted from 61% to a much more respectable 27%, while its accuracy edged up from 32% to 34%. Its attempt rate also fell from 77% to 53%, indicating a more refined and less speculative approach to tasks.

User Experience and Pricing Considerations

While Inkling 2 Small brings powerful capabilities, it’s important for prospective users to consider some of its practical aspects. Artificial Analysis characterizes the model as "notably slow per task" and "very verbose." For applications where speed and conciseness are paramount, these traits could be a drawback. Developers integrating the model into real-time systems or those requiring succinct responses will need to factor this into their planning and potentially implement additional processing layers to refine outputs.

From a cost perspective, Inkling 2 Small is priced at $0.30 per one million input tokens. Artificial Analysis deems this "somewhat expensive," noting that the median cost for models in its class is $0.24. While the price difference might seem marginal, for high-volume users, this could accumulate, making it a point of consideration when evaluating total cost of ownership against alternatives.

The Open-Weight Race: Comparing Inkling 2 Small

Inkling 2 Small enters a dynamic open-weight landscape. Its primary significance lies in its ability to challenge Chinese models that have previously led the pack. Models from DeepSeek and Moonshot have set a high bar, often rivaling closed models in capability. DeepSeek's R1, for instance, once performed similarly to OpenAI’s o1. Inkling 2 Small’s emergence provides a strong Western contender.

Another notable player in the Western open-weight space is Mistral’s Le Chonk, which, like Inkling 2 Small, was released via a staggered process (initial release followed by weights). Le Chonk has established itself as a top open model for coding, reportedly keeping pace with OpenAI’s GPT-6.1 Terra in this specific domain. While the article doesn't offer a direct comparative Artificial Analysis Intelligence Index score for Le Chonk against Inkling 2 Small, it highlights both models as strong Western players in specialized areas (coding for Le Chonk, broader utility including web/game design and coding for Inkling 2 Small).

Here’s a snapshot of how Inkling 2 Small compares on the Artificial Analysis Intelligence Index:

Model	Artificial Analysis Intelligence Index Score
Inkling 2 Small	31
Gemini 3.8 Flash	31
Gemini 4 Carbon	60
Claude Opus 5.5	59

This table clearly shows Inkling 2 Small holding its own against comparable open models like Gemini 3.8 Flash. However, it also illustrates the gap between current open-weight performance and the top-tier closed models represented by Gemini 4 Carbon and Claude Opus 5.5. The key takeaway for Inkling 2 Small is not necessarily beating the absolute top scorers, but rather offering a highly capable and accessible open-weight alternative that can significantly reduce reliance on expensive closed models, especially for coding-centric applications.

Buying Recommendation

Thinking Machines’ Inkling 2 Small is an excellent choice for developers, researchers, and organizations that prioritize powerful, open-weight AI solutions. If your projects involve complex coding, web design, or game development, and you appreciate the transparency and flexibility that come with an open-weight model, Inkling 2 Small is a strong contender. It's particularly well-suited for those looking to reduce dependence on costly closed models or seeking a competitive Western-developed alternative to dominant Chinese offerings.

However, potential users should be mindful of its reported verbosity, slower processing speed per task, and slightly higher pricing compared to the class median. For applications demanding extremely rapid or concise outputs, or those with very tight budgets, these factors warrant careful consideration. Ultimately, for those willing to work with its quirks, Inkling 2 Small offers a high-intelligence, open-source platform with significant gains over its predecessors, poised to elevate the capabilities of many AI-driven projects.

FAQ

Q: What are the main strengths of Inkling 2 Small?

A: Inkling 2 Small's main strengths lie in its strong coding capabilities, suitability for web and game design, support for image inputs with text outputs, and significant improvements in automation and accuracy compared to its previous version. It's also a leading Western open-weight model.

Q: Are there any downsides to using Inkling 2 Small?

A: Yes, according to Artificial Analysis, Inkling 2 Small is notably slow per task and can be very verbose in its outputs. Its pricing is also considered somewhat expensive at $0.30 per one million input tokens, exceeding the class median.

Q: How does Inkling 2 Small compare to other popular AI models?

A: Inkling 2 Small scored 31 on the Artificial Analysis Intelligence Index, tying with Gemini 3.8 Flash and positioning it as a top Western open-weight model. While not as high-scoring as premium closed models like Gemini 4 Carbon (60) or Claude Opus 5.5 (59), it offers a competitive open-source alternative, especially in coding, where it seeks to rival leading Chinese open models and other Western alternatives like Mistral's Le Chonk.