Artificial Analysis says Inkling 2 Small matches Gemini 3.8 Flash (high) in its index and would be the top open model outside China once its weights are out

The cost to run advanced AI agentic tasks decreases significantly as Inkling 2 Small’s weights are released.

Thinking Machines has launched Inkling 2 Small, a proprietary model that scored 31 on the Artificial Analysis Intelligence Index. This performance matches Google DeepMind's Gemini 3.8 Flash (high) and surpasses Xiaomi's MiMo-V2.6-Pro. The model shows significant improvements over its predecessor, Inkling Small, particularly in agentic tasks like AutomationBench-AA, FrontierCode 2.0, and HLE-Diamond. While Inkling 2 Small demonstrates different coding strengths compared to Claude Sonnet 5.5, it excels in creative agentic tasks and video generation using code. The model also exhibits reduced hallucination rates and significantly increased output token usage per task. Thinking Machines plans to release Inkling 2 Small's weights at a later date, which would position it as the top-scoring open-weights model developed outside China. The model's cost per task is higher than Inkling Small but competitive with other models of similar performance.

The planned open-weight release of Inkling 2 Small positions it as a strong contender against proprietary models, potentially democratizing access to high-performance AI for developers globally. Its significant gains in agentic tasks and reduced hallucination rates suggest a maturation in AI's ability to handle complex, multi-step workflows, impacting fields like software development, content creation, and business automation. The model’s performance benchmarks against major players like Gemini and Claude indicate a rapidly intensifying competitive landscape among AI providers, both open and closed.

This release signifies a strategic move by Thinking Machines to balance proprietary offerings with future open-source contributions, aiming to capture market share across different user segments. The substantial increase in output tokens per task, coupled with competitive pricing for its performance tier, could drive adoption for applications requiring extensive data processing or generation. It also highlights the ongoing trend of specialized AI models excelling in specific niches, pushing the boundaries of what was previously achievable with more general-purpose models.