Skip to content
Model infoModelsSnapshot: 15 Oct 2026

Baichuan AI / Baichuan medical research

Baichuan-M4

Medical agent research system with a trained reasoning core, persistent patient memory, retrieval and specialist visual tools.

Document extraction / OCR
restrictedClosed weightsText generationReasoningAgentsImage analysisDocument extraction
Official documentation (opens in a new tab)Reviewed 14 Oct 2026
6 of 17 specification fields collected
Context windowNot collectedAwaiting source review
Maximum outputNot collectedAwaiting source review
Input billingNot collectedBaichuan native platform
Output billingNot collectedNo inferred price

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit
Not collected
Maximum output
Not collected
Input modalities
Text, Image, Medical documentsSource for Input modalities (opens in a new tab)

System-level input through specialist tools, not evidence that the reasoning core natively accepts image tokens.

Output modalities
TextSource for Output modalities (opens in a new tab)
Reasoning
SupportedSource for Reasoning (opens in a new tab)
Tool calling
SupportedSource for Tool calling (opens in a new tab)

Agent-internal clinical tools; no public tools API contract established.

Structured output
Not collected
Streaming
Not collected
Batch processing
Not collected
Fine-tuning
Not collected
Knowledge cutoff
Not collected
Release date
Not collected
Parameters
Not collected
Architecture
Reasoning core plus Baichuan-Harness and clinical perception, retrieval and patient-memory toolsSource for Architecture (opens in a new tab)
Languages
Not collected
Context caching
Not collected
Published training method
Continuous-care reinforcement learning using SPAR++, reasoning-path compression, curriculum learning, SAPO and route replaySource for Published training method (opens in a new tab)

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

Baichuan native platform

Provider source (opens in a new tab)
API model ID
Not collected

No exact public API selector established.

Access
Named research system and homepage benchmark; absent from the reviewed public API selector list. Current customer access is unverified.
Regions
Native China platform. No additional regional endpoint or residency guarantee inferred.

Endpoint pricing has not been collected. Check the provider for applicable rates.

Data & enterprise terms

Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.

Caching

Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.

01 / INITIAL INPUTProcess & retain

The system processes eligible input and retains reusable state.

02 / LATER REQUESTCheck available state

Reuse depends on matching input, session state and the cache policy.

03 / CACHE HITReuse & generate

Reusing state can reduce repeated work. Any charges follow the billing rules below.

Baichuan native platform

Not verified

No model-specific prompt-cache contract or price verified.

How it works
Not collected

No model-specific prompt-cache contract or price verified.

Minimum reusable input
Not collected

No model-specific prompt-cache contract or price verified.

Lifetime & refresh
Not collected

No model-specific prompt-cache contract or price verified.

How to enable it
Not collected

No model-specific prompt-cache contract or price verified.

Usage fields
Not collected

No model-specific prompt-cache contract or price verified.

Billing rules
Not collected

No model-specific prompt-cache contract or price verified.

Limits & data boundaries
Not collected

No model-specific prompt-cache contract or price verified.

No verified cache rate for this endpoint. An empty rate is not a zero charge.

What the measurements say

Medical hallucination rate

Baichuan AI · provider-reported
3.3%

Provider-reported M4 system evaluation in arXiv 2606.08982v1. Section 5.2.1; lower is better. Provider evaluation, not a guarantee of clinical safety.

Evaluation source (opens in a new tab)

Long-context clinical memory

Baichuan AI · provider-reported
86.9score

Provider-reported M4 system evaluation in arXiv 2606.08982v1. Section 5.2.2. Evaluation score, not token capacity.

Evaluation source (opens in a new tab)

Baichuan-EBM citation precision

Baichuan AI · provider-reported
90score

Provider-reported M4 system evaluation in arXiv 2606.08982v1. Section 5.3.1; retrieval evaluation on 657 expert-annotated questions.

Evaluation source (opens in a new tab)

Medical OCR unstructured similarity

Baichuan AI · provider-reported
0.838score

Provider-reported M4 system evaluation in arXiv 2606.08982v1. Table 3, system perception tools; higher is better.

Evaluation source (opens in a new tab)

Medical OCR structured matching

Baichuan AI · provider-reported
0.914score

Provider-reported M4 system evaluation in arXiv 2606.08982v1. Table 3, system perception tools; higher is better.

Evaluation source (opens in a new tab)

X-ray/pathology CIDEr macro-average

Baichuan AI · provider-reported
0.1892score

Provider-reported M4 system evaluation in arXiv 2606.08982v1. Table 4, seven public benchmarks; specialist report-generation tool.

Evaluation source (opens in a new tab)

X-ray/pathology GREEN-LLM macro-average

Baichuan AI · provider-reported
0.8435score

Provider-reported M4 system evaluation in arXiv 2606.08982v1. Table 4, seven public benchmarks; model-judged metric.

Evaluation source (opens in a new tab)

f17k TOP-1 EXACT

Baichuan AI · provider-reported
30.78%

Provider-reported M4 system evaluation in arXiv 2606.08982v1. Table 5, 4,893 dermatology cases and evidence-driven tool use.

Evaluation source (opens in a new tab)

f17k TOP-1 CATEGORY

Baichuan AI · provider-reported
39.53%

Provider-reported M4 system evaluation in arXiv 2606.08982v1. Table 5, coarser category metric, not exact diagnosis.

Evaluation source (opens in a new tab)

f17k TOP-1 BLOCK

Baichuan AI · provider-reported
42.71%

Provider-reported M4 system evaluation in arXiv 2606.08982v1. Table 5, coarser block metric, not exact diagnosis.

Evaluation source (opens in a new tab)

Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.

The record behind the model

Why it is included

The original report describes a newly trained M4 reasoning core and integrated medical agent system. No exact M4 weights appear in the reviewed developer inventory. Closed classification is an inference for the named research release, not a claim that its vision tools or open M3 predecessor are closed-weight.

Classification evidence (opens in a new tab)

Lifecycle dates

Research report submitted8 Jun 2026Source (opens in a new tab)

Sources reviewed