The 2026 World Artificial Intelligence Conference (WAIC) opened today, and on its first day, Baichuan Intelligence showcased the initial real-world applications of its Baichuan-M4 model in specialized fields such as oncology and pediatrics.

These applications include an oncology-specific AI system developed in collaboration with the National Cancer Center and the Cancer Hospital of the Chinese Academy of Medical Sciences, as well as an AI pediatrician developed with the National Children's Medical Center and Beijing Children's Hospital. Both systems are currently undergoing validation in actual clinical settings.

Developed jointly by Baichuan Intelligence and a research team from Tsinghua University, Baichuan-M4 offers medical enhancement capabilities such as low hallucination rates, strong evidence-based reasoning, deep conversational follow-up, and long-term memory for both doctors and individual users.

These capabilities underpin the "Baixiao Yi" service for family health and are also being clinically validated in specialized scenarios like oncology and pediatrics.

According to public evaluation results, M4 achieved a top global score of 68.6 on the authoritative medical benchmark HealthBench, proposed by OpenAI, leading the second-place GPT-5.5 by over 10 points.

On the challenging "Hard" subset, which tests complex clinical decision-making, M4's lead extended to 15.9 points.

Furthermore, M4 demonstrated a factual hallucination rate of just 3.3%, the lowest in the industry. In comparison, GPT-5.5's rate was 3.8%, Claude Opus 4.7's was 6.9%, and DeepSeek-V4-Pro's was 9.8%.

In the evidence-based medicine framework, a doctor's trust in a conclusion relies not on how "doctor-like" it sounds, but on the authority, accuracy, and traceability of the evidence behind it.

To this end, Baichuan pioneered an "evidence anchoring" technology, ensuring that every medical conclusion generated by the model can be precisely linked to specific paragraphs in original research papers or guidelines.

Additionally, by adhering to a six-source evidence-based paradigm, the model retrieves information exclusively from authoritative medical sources, avoiding data scraping from the open web.

On the Baichuan-EBM benchmark for evidence-based medicine, M4 achieved an evidence citation accuracy of 90.0, significantly outperforming similar products.

M4 also possesses the deep conversational follow-up and extensive memory capabilities required in clinical practice, enabling it to manage a patient's complete medical history from initial consultation through multiple follow-ups.

When combined with Harness scheduling, memory, and self-evolution capabilities, the model has evolved from a powerful medical brain into a medical agent capable of conducting continuous diagnosis and treatment.

Currently, doctors in top-tier hospitals in China work over 9.5 hours daily on average, and primary care family doctors are each responsible for approximately 2,000 registered residents, highlighting the ongoing risk of insufficient high-quality medical resources.

Leveraging M4, Baichuan introduced "Baixiao Yi," an AI family doctor designed for comprehensive disease management throughout the entire patient journey.

This system not only helps patients understand their medical consultations but also proactively manages the health of entire families. The app handles symptom inquiry, pre- and post-consultation preparation, condition analysis, and medical advice interpretation for each visit.

Meanwhile, the WeChat BOT extends into users' daily lives, providing continuous proactive health management and follow-up care.

At the WAIC exhibition, Baichuan also showcased a new "Oncology AI Assistant." Addressing the challenges of information overload and complex decision-making in cancer treatment, this assistant, grounded in evidence-based medicine, integrates the latest guidelines, clinical trials, and real-world research in real-time.

It deconstructs medical records into clinical questions and dynamically retrieves evidence from multiple sources, enabling precise, second-level indexing.

For patients and their families who often struggle to discern the differences between treatment recommendations from multiple specialists or to understand the benefits and risks of various options, the AI oncologist can break down each proposal in detail.

It helps patients see the underlying evidence, survival benefits, and potential adverse reactions, facilitating more informed discussions with their doctors.

Trials have shown that the Oncology AI Assistant and the AI oncologist achieved 100% consistency with multidisciplinary team (MDT) expert opinions on guideline-based treatment plans, and a 94% consistency rate on complex cases outside standard guidelines.