dots3-note-prev: dots3-note: Ranked #1 among models under 500B, achieving 79.1% on MMMU-Pro

This is a preview version released from the dots3-note model lineup by dots-studio. It is classified as a multimodal LLM that processes image, audio, and video inputs alongside text generation. Distributed under the Apache-2.0 license, it is available for use in commercial environments.

It highlights long-context processing and agentic task execution as key features. It supports a conversational interface and allows automation of workflows that execute external APIs or functions through tool calling capabilities. It is specialized for code writing and problem-solving tasks that require complex reasoning.

According to published benchmark results, it recorded scores of 78.4% on SWE-bench Verified and 61% on SWE-bench Pro. It achieved 79.1% on MMMU-Pro, which measures multimodal understanding, ranking first among models with fewer than 500B parameters. Additionally, it showed top-tier performance among comparable models with a comprehensive score of 61.7% on WildClawBench.

It is suitable for developers planning software engineering automation, complex multimodal data analysis, and building tool-based agent systems. Given its verified performance in code modification and bug fixing (SWE-bench), it has value for use in CI/CD pipelines or code review assistance tools. You can run inference directly via the Transformers library or serve it as an API through Hugging Face Inference Endpoints.