Updated 45 min ago · first seen 11 Sept 2026
model_01M296QTDT4HJPWMMAAZ4M4VQD
- Parameters
- 3.8B
- T2 · 1 h ago
- Released
- 11 Aug 2026
- T2 · 1 h ago
Specification
- Release date
- 11 Aug 2026
Source:Microsoft Research — publications & blogT2observed 1 h agomediumLLM-extracted
- Status
- announced
Source:Microsoft Research — publications & blogT2observed 1 h agomediumLLM-extracted
- Openness
- proprietary
Source:Microsoft Research — publications & blogT2observed 1 h agomediumLLM-extracted
- Architecture
- SigLIP2-so400M vision encoder and Phi-4-mini-instruct (3.8B) language model connected through a lightweight adapter, with task-specific auxiliary heads for classification and visual grounding
Source:Microsoft Research — publications & blogT2observed 1 h agomediumLLM-extracted
- Parameters
- 3.8B
Source:Microsoft Research — publications & blogT2observed 1 h agomediumLLM-extracted
- Modalities
- image, text
Source:Microsoft Research — publications & blogT2observed 1 h agomediumLLM-extracted
- Input modalities
- image, text
Source:Microsoft Research — publications & blogT2observed 1 h agomediumLLM-extracted
- Output modalities
- text
Source:Microsoft Research — publications & blogT2observed 1 h agomediumLLM-extracted
Each value shows its source, tier and observation time. Conflicting claims are kept side by side and flagged — never averaged. How AI Atlas records facts →
Provenance
Attributed facts
15
Source tiers
T215
Freshest observation
1 h ago
Conflicts
None
Modalities
- Modalities
- imagetext
- Input
- imagetext
- Output
- text
Capabilities
Tool calling
Yes
Microsoft Research — publications & blog · T2
Structured output
Yes
Microsoft Research — publications & blog · T2
Reasoning
Yes
Microsoft Research — publications & blog · T2
Vision
Yes
Microsoft Research — publications & blog · T2
Audio
No
Microsoft Research — publications & blog · T2
Fine-tuning available
Unavailable
No structured attributes yet.
No benchmark results recorded
Current prices
No current prices recorded
Price history
Memory need = bytes per parameter (4-bit ≈ 0.5 × 1.15 overhead, 8-bit 1.0, fp16 2.0) + a KV-cache allowance. Not a measurement.
| Hardware | Quantization | Memory | Est. need | Fits |
|---|---|---|---|---|
| Apple M3 Ultra | 4bit | — | 2.7 GB | Yes |
| AMD Instinct MI325X | 4bit | 256 GB | 2.7 GB | Yes |
| AMD Instinct MI300X | 4bit | 192 GB | 2.7 GB | Yes |
| Apple M2 Ultra | 4bit | — | 2.7 GB | Yes |
| NVIDIA B200 | 4bit | 180 GB | 2.7 GB | Yes |
| NVIDIA H200 | 4bit | 141 GB | 2.7 GB | Yes |
| Apple M1 Ultra | 4bit | — | 2.7 GB | Yes |
| Apple M3 Max | 4bit | — | 2.7 GB | Yes |
| Apple M4 Max | 4bit | — | 2.7 GB | Yes |
| NVIDIA DGX Spark | 4bit | 128 GB | 2.7 GB | Yes |
| Apple M2 Max | 4bit | — | 2.7 GB | Yes |
| NVIDIA A100 80GB | 4bit | 80 GB | 2.7 GB | Yes |
| NVIDIA H100 SXM | 4bit | 80 GB | 2.7 GB | Yes |
| Apple M1 Max | 4bit | — | 2.7 GB | Yes |
| Apple M4 Pro | 4bit | — | 2.7 GB | Yes |
| Apple M3 Pro | 4bit | — | 2.7 GB | Yes |
| Apple M1 Pro | 4bit | — | 2.7 GB | Yes |
| Apple M2 Pro | 4bit | — | 2.7 GB | Yes |
| Apple M4 | 4bit | — | 2.7 GB | Yes |
| NVIDIA GeForce RTX 5090 | 4bit | 32 GB | 2.7 GB | Yes |
| Apple M2 | 4bit | — | 2.7 GB | Yes |
| Apple M3 | 4bit | — | 2.7 GB | Yes |
| NVIDIA GeForce RTX 3090 | 4bit | 24 GB | 2.7 GB | Yes |
| NVIDIA GeForce RTX 4090 | 4bit | 24 GB | 2.7 GB | Yes |
| Apple M1 | 4bit | — | 2.7 GB | Yes |
Ancestors 1
This model
CARE-X
3.8B params
Quantizations 0
None recorded.
Descendants 0
None recorded.
Papers 0
No papers linked yet.
Repositories 0
No repositories linked yet.
As of
Rewind the record: see this entity's attributes exactly as AI Atlas knew them on a given day.
Claim history
Release daterelease_date1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| 11 Aug 2026 | → current | current | Microsoft Research — publications & blogT2 | medium | llm |
Statusstatus1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| announced | → current | current | Microsoft Research — publications & blogT2 | medium | llm |
Opennessopenness1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| proprietary | → current | current | Microsoft Research — publications & blogT2 | medium | llm |
Architecturearchitecture1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| SigLIP2-so400M vision encoder and Phi-4-mini-instruct (3.8B) language model connected through a lightweight adapter, with task-specific auxiliary heads for classification and visual grounding | → current | current | Microsoft Research — publications & blogT2 | medium | llm |
Parametersparameter_count1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| 3.8B | → current | current | Microsoft Research — publications & blogT2 | medium | llm |
Modalitiesmodalities1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| image, text | → current | current | Microsoft Research — publications & blogT2 | medium | llm |
Input modalitiesmodalities_input1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| image, text | → current | current | Microsoft Research — publications & blogT2 | medium | llm |
Output modalitiesmodalities_output1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| text | → current | current | Microsoft Research — publications & blogT2 | medium | llm |
Audioaudio1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| No | → current | current | Microsoft Research — publications & blogT2 | medium | llm |
Reasoningreasoning1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| Yes | → current | current | Microsoft Research — publications & blogT2 | medium | llm |
Safety notessafety_notes1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| CARE-X is a research model and not a Microsoft product offering or medical device. It has not been cleared or approved by any regulatory authority and is not intended for clinical diagnosis, screening, or patient care. The results described below are retrospective research findings and do not establish the safety, effectiveness, or suitability of CARE-X for any clinical use. | → current | current | Microsoft Research — publications & blogT2 | medium | llm |
Structured outputstructured_output1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| Yes | → current | current | Microsoft Research — publications & blogT2 | medium | llm |
Tool callingtool_calling1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| Yes | → current | current | Microsoft Research — publications & blogT2 | medium | llm |
Training data notestraining_data_notes1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| Trained on real-world Indian clinical data from Narayana Health, including rare ICU pathologies and CT-confirmed enlargement conditions. Used de-identified, retrospective clinical data under applicable institutional ethics review. | → current | current | Microsoft Research — publications & blogT2 | medium | llm |
Visionvision1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| Yes | → current | current | Microsoft Research — publications & blogT2 | medium | llm |
Claims are temporal and append-only: a new observation closes the previous claim (valid_to) instead of overwriting it. Conflicting claims from different sources are kept side by side and flagged — never averaged. Methodology →
| Source | Document | Type | Tier | Last observed | Snapshots |
|---|---|---|---|---|---|
| Microsoft Research — publications & blog | microsoft.com/en-us/research/blog/introducing-care-x-towards-clinically-useful-radiology-vlms-with-auxiliary-supervision-reward-aligned-learning-and-tool-augmented-measurement | news | T1· Official | 1 h ago | 1 |
Tier 1 = official/primary, 2 = quality secondary, 3 = community, 4 = unverified. Every snapshot is archived; see all sources and the methodology.