HakkoAI is powered by LynkSoul VLM v1—a visual language model developed by our team and trained on massive amounts of gaming data. This allows the AI to truly see and understand gameplay scenarios with exceptional visual awareness. Thanks to its highly specialized training, LynkSoul VLM v1 stands as the most effective visual language model built for gaming.
Expertise
In League of Legends testing scenarios, we constructed an image test set containing 500+ samples, covering the vast majority of key scenarios. Results demonstrate that the LynkSoul VLM v1 significantly outperforms general vision models in recognition accuracy, suggestion practicality, and naturalness of language expression.


Generalization
To evaluate the generalization ability of our model, we built a test dataset containing multiple games that were excluded from training. Given the limited availability of online data for some titles, we intentionally selected such data-scarce games to assess the model’s performance on unseen content.
Experimental results show that LynkSoul-VLM-v1 achieves strong generalization performance. Across three key metrics—visual understanding accuracy, gameplay comprehension, and natural language expression—it consistently outperforms baseline models, including OpenAI-4o-Latest, Claude-4-Sonnet, and Gemini-2.5-Flash.
The findings confirm that LynkSoul-VLM-v1 exhibits robust generalization in low-resource gaming environments and demonstrates effective cross-domain knowledge transfer.


Speed&Cost
In addition to accuracy, we equally focus on inference speed. Due to the dynamic path selection of the Hybrid Model System, LynkSoul-VLM-v1 demonstrates significant advantages in inference speed.
Inference Latency Comparison Results:






