Inkling by @thinkymachines has debuted at #9 among open-weight models and #30 overall in the Agent Arena, marking it as the best open model from the US for long-running agentic tasks and the only US model in the top 10. While Inkling demonstrates strength in recovering from command-line errors, ranking #16 in that category, it faces challenges with user sentiment, as reflected in its low praise-to-complaint ratio of -18.0%. The performance of open-weight models like Inkling is critical in the context of US AI development, where such models are increasingly competing on international benchmarks for their capabilities in complex, multi-tool workflows.

arena: Agent Arena is a benchmarking platform that assesses AI models using millions of real-world long-horizon agentic tasks. It provides rankings for both open-weight and overall models across domains such as code generation and research workflows. The platform incorporates tools like web search, file systems, and terminals to evaluate model performance.
Inkling: Inkling is an open-weight AI model developed by Thinky Machines focused on agentic capabilities. It made a notable entry into the Agent Arena evaluation platform measuring performance across complex workflows. The model stands out as the top US-based entry in the open-weight category for long-running agent tasks.
Thinky Machines: Thinky Machines is the development team responsible for creating the Inkling AI model. The team achieved a strong initial ranking for their model in the Agent Arena leaderboard for open-weight systems. Their work highlights advancements in US-based open models for practical agentic applications.

Error Recovery: Agent benchmarks increasingly test AI models on their ability to recover from command-line errors during extended tasks.
US AI Development: Open-weight models developed in the US are positioning competitively in international benchmarks for agent capabilities.
AI Model Evaluation: Leaderboards for agentic tasks are highlighting strengths of open-weight models in real-world workflows involving multiple tools.