XunTu Closes RMB 300M Round; Molecular Mind Publishes in Science Advances; YuanKong AI Enters HP Workstations

XunTu Technology said the B round was joined by China Insurance Investment, a national investment platform of the insurance industry, and GF Xinde, the private-equity fund subsidiary of GF Securities. Qiming Venture Partners, Amber Capital, Jiacheng Capital, Xinchen Capital, Zhongding Capital and existing shareholders also participated. Huaxing Capital continued as exclusive financial adviser. The company said it will increase investment in models and products, expand multi-asset clients, deepen industry collaboration and steadily advance overseas expansion while focusing on its home market.

The company said China Insurance Investment's participation gives it a foothold in insurance research and allocation, where investment chains span equities, fixed income and alternatives and face strict compliance requirements. GF Xinde's fund focuses on the investment advisory chain, including independent advisory teams, service providers and financial technology companies. XunTu's AlphaPai AI research assistant had more than 120,000 professional users and served over 8,000 financial and asset management institutions, with website traffic among the top tier of comparable global products, according to the company. In April 2026, it launched PaiWork, an AI research workstation that integrates financial data, research tools, office suites and knowledge bases. Since launch, AI interactions and token usage have grown dozens of times, and institutional repurchase rates have continued to rise, the company said. In June 2026, it introduced the iRaB research evaluation system and the ORE research ecosystem partner program with more than 40 securities research institutes and data service providers. Founder Li Luodan said serious-scenario AI is closer to high-end manufacturing and requires long-term accumulation, with competitiveness coming from many small gains that form a hard-to-copy systemic advantage.

Molecular Mind said its study on QuantaMind was published in Science Advances. The platform is described as the first to push reactive molecular dynamics with accuracy close to density functional theory to biological systems with tens of thousands of atoms and time scales of tens of nanoseconds. Extended tests after paper submission reached 100,000 atoms and 100 nanoseconds. In the paper, the largest validated system was a 24,001-atom water system and the longest simulation was 20 nanoseconds. For a 17,792-atom PET hydrolase, QuantaMind ran a 6-nanosecond reactive molecular dynamics simulation that covered a complete catalytic cycle.

On an NVIDIA A100 80GB GPU, QuantaMind's inference speed reached 4.2×10⁻⁶ seconds per atom per step, comparable to state-of-the-art machine learning force fields, according to the paper. For an approximately 18,000-atom PETase system, each nanosecond of simulation took about 20 hours. The latest tests reached 2.1×10⁻⁶ seconds per atom per step, reducing a 100,000-atom reactive system's simulation time to 0.25 seconds per time step. The team extracted 342 configurations from reaction trajectories for independent DFT single-point calculations and found Pearson correlation coefficients above 0.99 for atomic force components. In a 24,001-atom water system, simulated proton diffusion coefficient was 0.87±0.10 Ų/ps, compared with experimental values of 0.94/0.96. For a 9,999-atom water system simulated for 3 nanoseconds, captured self-ionization events gave a pH of 6.34±0.06, close to the experimental value of 6.13±0.03. A 20-nanosecond simulation of lysozyme in a buffer estimated the pKa of histidine residue 15 at about 5.81, compared with 5.5 measured by nuclear magnetic resonance.

In industrial applications, Molecular Mind said QuantaMind helped design a pH-sensitive antibody whose candidate showed a dissociation rate at pH 6.0 that was 62 times that at pH 7.4. In enzyme engineering, the company used QuantaMind to identify reaction mechanisms and key mutation sites for complex catalytic enzymes, narrowing wet-lab testing. Founder and corresponding author Xu Jinbo said AI had learned to answer what molecules look like, and the next step is to answer why they move and react as they do. QuantaMind is intended to connect molecular design, mechanism simulation and experimental validation and will be integrated into Molecular Mind's MoleculeOS.

YuanKong AI released its on-device model Boxer, office agent YuanKong AI Work and research agent YuanKong AI Science. The company defines productivity as YuanKong on-device AI equals model times Agent times device. It argues that growth from static internet data is peaking, citing Stanford's AI Index 2026 reference to Peak Data and a JMLR study that ran more than 400 training runs, up to 9 billion parameters and 900 billion training tokens, and found diminishing returns as existing data was repeated. YuanKong says devices can generate scarce data through real tasks and feedback, allowing models, agents and devices to evolve together.

Boxer is positioned in the mid-size 20B to 100B parameter range. In YuanKong's WorkArena evaluation, Boxer-35B-A3B-v0-0819 scored 73.1, first among comparison groups and 10.3 points ahead of the second, with some tasks surpassing GPT-4o from more than a year earlier, according to the company. On 40 TOPS of compute and less than 8GB of memory, Boxer retained 87.3% of the original model's task performance. YuanKong AI Work connects goals, cross-application execution, result verification, state updates and continued work, and can run locally and offline. YuanKong AI Science extends from literature research, data analysis, bioinformatics and protein engineering to experimental equipment, linking experiment design, execution and data return. A Memory system stores task trajectories, environmental events and user feedback as task states, knowledge assets and preferences so agents can resume state, reuse knowledge and learn preferences across tasks, devices and time.

YuanKong said it has worked with chip and hardware makers and had commercial orders with HP last year. It recently signed a strategic cooperation memorandum with HP. YuanKong AI Work and the 35B on-device model have been deeply adapted to HP Z series ZGX Nano G1n AI workstation, ZBook 8 G2i mobile workstation and Z2 Mini G1a desktop workstation, and are being offered to manufacturing, finance, healthcare and research sectors. The company plans to follow HP SKUs into notebooks, workstations and servers and use HP's global channels to expand overseas.