This model was completed using a private model-merging algorithm. Its real-world performance is significantly stronger than the original models, and mathematically, it forms an almost perfect centroid.

In private agent usage, the experience is nothing short of exceptional. Aside from not yet surpassing GPT-5.5 in raw intelligence, there are virtually no major shortcomings.

The overall experience is already extremely close to GPT-5.5.

🧠 Research Highlight

This work demonstrates the successful merging of quantized models.

The resulting model quality exceeds that of the same model sources under BF16, showing that quantized-model merging can produce results beyond the original BF16-level model composition.

🙏 Credits

Special thanks to the authors of the following models:

Qwen/Qwen-AgentWorld-35B-A3B
infly/Infinity-Parser2-Pro
InternScience/Agents-A1
huihui-ai/Huihui-Nex-N2-mini-abliterated
Hcompany/Holo-3.1-35B-A3B
Gryphe/WorldSim-Opus-3.6-35B-A3B
AEON-7/Ornith-1.0-35B-AEON-Ultimate-Uncensored-BF16

💖 Open Sponsorship

Support this research and future model development:

Ko-fi
PayPal