MM-ToolSandbox (partial credit)

MM-ToolSandbox (partial credit) is a benchmark for agents that perceive and act across visual interfaces. The official description says: "MM-ToolSandbox (partial credit) as evaluated on the LFM3-VL-2B card (vision benchmark, with-thinking setting): Agentic category, with-thinking column". Below we describe in simple words what benchmarks of this type generally examine.

MM-ToolSandbox (partial credit) measures agents that perceive and act across visual interfaces. In detail, it checks:

• interpreting what is on screen and deciding what to do next
• combining visual perception with tool or UI actions
• completing multi-step tasks that require looking and acting alternately
• recovering when a UI action does not produce the expected result

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.

Official methodology note: MM-ToolSandbox (partial credit) as evaluated on the LFM3-VL-2B card (vision benchmark, with-thinking setting): Agentic category, with-thinking column. Multimodal MM-ToolSandbox scored with partial credit; inventory 'ToolSandbox' (slug toolsandbox) is presumed text-based, so kept separate to avoid mis-assignment. Value equals the Agentic category average (single benchmark).

What is the MM-ToolSandbox (partial credit) benchmark?

MM-ToolSandbox (partial credit) is a benchmark for agents that perceive and act across visual interfaces. The official description says: 'MM-ToolSandbox (partial credit) as evaluated on the LFM3-VL-2B card (vision benchmark, with-thinking setting): Agentic category, with-thinking column'. Below we describe in simple words what benchmarks of this type generally examine.

What does MM-ToolSandbox (partial credit) test?

MM-ToolSandbox (partial credit) mainly measures agents that perceive and act across visual interfaces. Concretely, it checks: interpreting what is on screen and deciding what to do next; combining visual perception with tool or UI actions; completing multi-step tasks that require looking and acting alternately.

How is the MM-ToolSandbox (partial credit) score calculated?

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better. The declared maximum score of this benchmark is 100.0.

Where do I get MM-ToolSandbox (partial credit) data and scores?

The official benchmark page/reference is linked on this page. Every score on Open LLM Radar is also available as open data: download the scores JSON (http://open-llm-radar-api:8000/api/v1/benchmarks/mm-toolsandbox/scores) or pull it directly from the Open LLM Radar API with any HTTP client.