APMIC (Accelerate Private Machine Intelligence Company), a leader in enterprise AI autonomy solutions, announced today (19th) that its enterprise-grade fine-tuning and distillation solution “PrivModel” now fully supports OpenAI’s gpt-oss series models. In specific application scenarios, accuracy can be improved by approximately 40% while inference costs are reduced by 90%. In addition, APMIC announced the appointment of senior GPU and AI chip expert Ian Chen, who has more than 20 years of industry experience, as President of U.S. operations. He will be responsible for expanding the North American market and strategic partnerships, combining technology upgrades with localized teams to accelerate the secure deployment and business value realization of privatized AI worldwide.

gpt-oss Enhanced by “PrivModel”: Accuracy Leaps, Costs Optimized

APMIC has upgraded its proprietary “S1 Distillation” into “PrivModel” to support OpenAI’s newly released gpt-oss series, including gpt-oss-120b and gpt-oss-20b, raising enterprise solutions to multimodal fine-tuning. According to APMIC’s internal testing, in specific enterprise applications, distilled models can improve gpt-oss accuracy by about 40% while maintaining FP4 precision, and simultaneously reduce inference costs by about 90%. The distilled models also run efficiently on RTX Pro 6000 platforms and support Traditional Chinese Chain-of-Thought (CoT) reasoning, enabling enterprises to achieve privatized deployment of Reasoning AI under regulatory compliance.

Further testing by APMIC revealed that OpenAI’s models have strong protective mechanisms, making traditional QLoRA quantization fine-tuning limited in learning domain-specific knowledge. Successful fine-tuning requires a combination of CPT, SFT, and RL. In TMMLU+ (Taiwan Traditional Chinese Knowledge Benchmark), all gpt-oss models scored above 80 (compared to gpt-3’s below-60 in 2023). In healthcare, they outperformed o4 mini by about 13 points, while in math and logic reasoning, they trailed by only 2 points. Building on this, the “PrivModel” solution further enhances gpt-oss with visual capabilities, enabling enterprises to efficiently and securely build lightweight, commercially valuable privatized AI models.

Jerry Wu, Founder and CEO of APMIC, stated: “OpenAI’s gpt-oss brings new momentum for enterprise AI adoption. After being fine-tuned and distilled through ‘PrivModel,’ and deployed on the ‘PrivStation’ framework, the models can match o4-mini levels while also possessing visual capabilities. This allows Taiwanese enterprises to safeguard their knowledge assets while achieving high-performance inference with 90% cost savings. This is not only about performance gains but also full implementation from data protection to real-world application. We collaborate closely with inference compute partners such as CKmates, GMI Cloud, Glows AI, and Ubilink, as well as hardware partners Dell, HPE, and NVIDIA, while leveraging CyCraft’s security protections to ensure safe model operation across the Blackwell architecture. This truly delivers an enterprise AI pathway ‘from model to decision.’”

Senior GPU Expert to Lead North American Market

APMIC also announced the appointment of Ian Chen as President of U.S. operations. With his expertise in GPU acceleration and AI chips, he will focus on driving North American market expansion and strategic partnerships, advancing the deployment of “PrivModel” and “PrivStation” applications. He will collaborate with GPU cloud providers, OEMs, and design software companies to expand AI adoption in industries such as finance and manufacturing that require compliance-ready AI.

Ian Chen brings over 20 years of industry and management experience. He joined NVIDIA in Silicon Valley in 2002 as a graphics chip application engineer, later holding engineering and business management roles in NVIDIA’s Hong Kong and Taiwan branches, where he oversaw high-end graphics card and data center GPU businesses. He holds a Master’s in Electrical Engineering from Stanford University and a Master’s in Business Analytics from Santa Clara University.

【關於 APMIC】

Founded in Taiwan in 2017, APMIC (Accelerate Private Machine Intelligence Company) is a leading provider of enterprise AI autonomy solutions. With natural language understanding (NLU) as its core technology, APMIC has developed the “PrivModel” fine-tuning and distillation service and the “PrivStation” one-stop privatization solution, enabling enterprises to upload data and build dedicated AI brains without coding. APMIC’s products have been adopted by over 1,100 enterprises and government agencies across healthcare, manufacturing, finance, and retail, committed to empowering every organization to build its own AI brain.：https://apmic.ai/