Nexusflow.ai, a startup dedicated to generative artificial intelligence network security, announced today the launch of NexusRaven-V2, an open-source generative AI model with 13 billion parameters. NexusRaven-V2 possesses the capability of function invocation and claims to outperform GPT-4 in terms of software tool usage. The function invocation capability allows this large-scale language model to comprehend human instructions and translate them into precise function/application interface calls for use in various software tools. This capability is the core of OpenAI's Assistant API and is crucial for enabling assistants and agents to use software tools. This new model is trained for instruction conversion from the CodeLlama-13B model and achieves a 7% higher success rate in tool usage compared to the latest OpenAI GPT-4 model in benchmark tests for general software tool usage planned by humans. The data curation pipeline of the new model combines open datasets and Meta's Llama ecosystem, curating a large amount of high-quality data used to train NexusRaven-V2. The data curation and training process of NexusRaven-V2 did not involve proprietary large-scale language models like OpenAI's GPT-4, allowing enterprise customers to own and utilize the model to build assistants and agents. Nexusflow emphasizes that this is particularly important for enterprise applications that require the latest information, quality, security, and deep domain customization. According to Nextflow, NexusRaven-V2 offers superior quality and has the potential to revolutionize workflow automation in complex software with smaller model sizes and higher accuracy. Nexusflow's customers can have their own generative AI models without the risk of using proprietary model vendor data. The company states that with the release of NexusRaven-V2, customers can enjoy superior generative AI quality, reliability, cost reduction, and security compared to solutions relying on proprietary model vendor APIs, while also enabling full and up-to-date customization, which is unpredictable in terms of quality degradation and knowledge lag. "NexusRaven-V2 surpasses OpenAI's GPT-4 model in terms of software tool usage," said Jiantao Jiao, co-founder and CEO, prior to the release. "This confirms Nexusflow's technical capability of providing solutions for enterprises utilizing open-source models." Nexusflow is a venture capital-backed startup, with its latest funding round occurring in September, raising $10.6 million. The funding was led by Point72 Ventures LP, with participation from Fusion Fund LLC and several individual investors.