InclusionMed launches medical AI evaluation platform, with a 22-model first index across 1,034 tasks
InclusionMed, which describes itself as the world's first independent evaluation and analysis platform for healthcare AI, has gone live with a nine-cell benchmark map (InclusionMed Atlas) and a first composite index (InclusionMed Index). The first round compares 22 models from 11 vendors across five benchmarks and 1,034 tasks.
InclusionMed, billed as the world's first independent evaluation and analysis platform for healthcare AI, has officially launched with two components: a benchmark map called InclusionMed Atlas and a first composite ranking called InclusionMed Index. The details come from a Chinese-language article published by QbitAI (量子位) on 2026-11-18, which the outlet reproduced with authorization from Ant Group (蚂蚁).
Who built it
The platform was built by researchers from 29 institutions, including Stanford University, Harvard University, Peking Union Medical College, Ant Group and InclusionMed.AI. The article says it is regarded as the healthcare counterpart to Artificial Analysis (AA), and that it fills a gap in comprehensive healthcare AI evaluation.
InclusionMed Atlas
Atlas organizes benchmarks along two dimensions: who the AI serves and what capability is tested.
- Roles: patients and the general public, medical professionals, healthcare administrators.
- Capabilities: knowledge, reasoning, task execution.
Crossing the three roles with the three capabilities produces a nine-cell grid. Each benchmark is placed according to its primary evaluation content. The stated purpose is to show which healthcare needs current evaluations cover and which still need to be filled in.
InclusionMed Index
The Index aggregates results from the included benchmarks to show each model's overall performance and its strengths and weaknesses across tasks.
The first round focuses on two areas: clinical diagnosis and treatment, and healthcare management. It covers:
- 5 benchmarks and 1,034 tasks
- 22 models from 11 vendors in China and abroad
- Independent retesting of results
The team selected benchmarks by weighing medical relevance, evaluation credibility and complementarity across scenarios. The first index consists of four public benchmarks plus the team's own InclusionMed-DxFiles. DxFiles is built from difficult cases published in top medical journals and tests whether a model can make clinical judgments from case clues.
The article does not name the four public benchmarks, the vendors or the models evaluated, and it gives no scores in the text.
Next steps
The project team says it is working with medical institutions and research teams to develop new evaluation tasks. Planned scenarios include shared doctor-patient decision-making in complex treatment choices and treatment scheduling under resource constraints. The aim is to assess AI in clinical care, patient decision support and healthcare management.
What this means
The article's starting point is that strong performance on health Q&A does not show that a model can handle specialized tasks such as spotting key clues in difficult cases, operating inside electronic health record systems, or completing multi-step research and operations work. It also says single-benchmark scores do not fully capture performance across medical tasks.
Atlas addresses this by mapping each benchmark to a role and a capability, so users can see coverage and gaps. The Index combines results into one comparison. Ant Group is among the contributing institutions and also supplied the article, so readers should treat the platform's independence claim as the organizers' own description. Per-model results and the identities of the 22 models are needed before the first index can be assessed.
Related Articles
Future Doctor's MedGPT beats 250 hospital physicians 88 to 12 on 40 complex lung cancer cases
Chinese medical AI company Future Doctor (未来医生) and thoracic surgery teams at top-tier hospitals report that its MedGPT won 88 to 12 against 250 physicians on 40 unpublished complex lung cancer cases. MedGPT scored 96.22 in a head-to-head comparison with general-purpose models, against 71.38 for GPT-5.4, 61.81 for Claude Opus 4.7 and 54.62 for Gemini 3.1 Pro.
Scale AI and IMDA launch SG-LegalClarify, testing 19 models on 1,860 Singapore law scenarios
Scale AI and Singapore's Infocomm Media Development Authority (IMDA) introduced SG-LegalClarify on November 12, 2026. The benchmark tests 19 language models on 1,860 scenarios across 10 domains of Singapore law, measuring both legal accuracy and whether models ask for missing facts before answering.
Databricks launches public AI Index: open-weight models reach 30.9% of enterprise tokens, up from 7.3% in July
Databricks has launched the AI Index, a public resource combining usage data from 10,000+ companies on Unity Gateway with weekly coding-agent evaluations. Open-weight models grew from 7.3% of enterprise token traffic in July to 30.9% in November, and cost per task varies far more than token prices suggest.
NVIDIA Vera Rubin NVL72 posts fastest time to train on six of seven MLPerf Training v6.1 benchmarks
In its first MLPerf Training appearance, NVIDIA's Vera Rubin NVL72 delivered the fastest time to train on six of seven benchmarks, with up to 2x the per-GPU performance of an optimized GB300 NVL72 baseline, according to NVIDIA. On DeepSeek-V3 671B it trained in 4.14 minutes on 1,024 GPUs versus 8.03 minutes for GB300 NVL72.
Comments
Loading...