Model reference — AI Medical · Medical images
Chest radiography is the highest-volume examination in most hospitals, which is why automated reading is further along here than anywhere else in imaging.
Three classes of model, three cost profiles. Classifiers return a probability per finding, run on modest hardware and are enough to order a queue. Vision-language models draft the findings text and cost considerably more per study. Image encoders produce no report at all — they turn a radiograph into a vector for retrieval, prior-study comparison, or a classifier trained on your own labels.
Every model on this page runs as part of a managed AI pipeline in our GPU clusters, with a dedicated private cluster in our cloud or an on-premise installation where medical governance requires it. Output is decision support for a qualified professional to review, not a diagnosis.
Each specification table gives three hardware tiers — Minimum, the smallest setup on which the model runs correctly; Medium, the usual production configuration; and High, a configuration sized for peak volume. The rate is what one server of that tier processes per hour. Use these figures for initial sizing only. Before production we benchmark your own data to confirm accuracy, latency, throughput and cost.
Vendor: Stanford AIMI
What it does: interprets a chest radiograph and drafts structured findings text, trained specifically for this examination rather than adapted from a general model.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | L40S 48 GB | A100 80 GB | H100 80 GB |
| VRAM | 48 GB | 80 GB | 160 GB |
| vCPUs | 16 | 32 | 64 |
| RAM | 64 GB | 128 GB | 256 GB |
| Server | 1× L40S 48 GB | 1× A100 SXM 80 GB | 2× H100 SXM 80 GB |
| Rate (studies/hour) | ≈ 400 | ≈ 1,500 | ≈ 4,200 |
Vendor: Microsoft Research
What it does: drafts a report in which each finding is grounded to the region of the image that supports it, so a reader can verify every statement.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | L40S 48 GB | A100 80 GB | H100 80 GB |
| VRAM | 48 GB | 80 GB | 160 GB |
| vCPUs | 16 | 32 | 64 |
| RAM | 64 GB | 128 GB | 256 GB |
| Server | 1× L40S 48 GB | 1× A100 SXM 80 GB | 2× H100 SXM 80 GB |
| Rate (studies/hour) | ≈ 250 | ≈ 900 | ≈ 2,600 |
Vendor: Microsoft Research
What it does: a smaller radiology vision-language model that drafts findings on a single mid-range GPU, which makes per-study cost practical at department volume.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3090 | L40S 48 GB | A100 80 GB |
| VRAM | 24 GB | 48 GB | 80 GB |
| vCPUs | 8 | 16 | 32 |
| RAM | 32 GB | 64 GB | 128 GB |
| Server | 1× RTX 3090 24 GB | 1× L40S 48 GB | 1× A100 SXM 80 GB |
| Rate (studies/hour) | ≈ 600 | ≈ 2,200 | ≈ 6,000 |
Vendor: Mila, Université de Montréal
What it does: returns probabilities for eighteen common findings. Light, well characterised across datasets and the dependable choice for triage ordering.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 8 GB | 24 GB | 80 GB |
| vCPUs | 6 | 12 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (studies/hour) | ≈ 9,000 | ≈ 32,000 | ≈ 90,000 |
Vendor: Microsoft Research
What it does: produces an image representation rather than text, for similarity search, retrieval of comparable prior studies, and training a classifier on your own labels.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3090 | L40S 48 GB | A100 80 GB |
| VRAM | 24 GB | 48 GB | 80 GB |
| vCPUs | 8 | 16 | 32 |
| RAM | 32 GB | 64 GB | 128 GB |
| Server | 1× RTX 3090 24 GB | 1× L40S 48 GB | 1× A100 SXM 80 GB |
| Rate (studies/hour) | ≈ 5,000 | ≈ 18,000 | ≈ 50,000 |
Vendor: Google DeepMind
What it does: a general medical vision-language model that answers questions about a radiograph and drafts text, useful where the same service also handles other image types.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3090 | L40S 48 GB | A100 80 GB |
| VRAM | 24 GB | 48 GB | 80 GB |
| vCPUs | 8 | 16 | 32 |
| RAM | 32 GB | 64 GB | 128 GB |
| Server | 1× RTX 3090 24 GB | 1× L40S 48 GB | 1× A100 SXM 80 GB |
| Rate (studies/hour) | ≈ 800 | ≈ 2,800 | ≈ 7,500 |
If the goal is a better-ordered worklist, a classifier does the job on a single mid-range card. If the goal is draft text, a vision-language model is required and the sizing changes accordingly. We benchmark the shortlist against studies your own readers have already reported, so the accuracy figures you plan around come from your population rather than a public leaderboard.
At the start of a project we run a short proof of concept on a sample of your own data, measuring the accuracy and the throughput the model actually achieves on your material. The estimates on this page are replaced with real figures, so the cost and the schedule for the full engagement are known before anything is committed.
Send a representative sample, your expected volume and your latency target for chest x-ray reporting support. We benchmark the shortlisted models, recommend the lowest-cost GPU configuration that meets the target, and scale it from pilot capacity to a dedicated production cluster — with the cost per unit of work known before you commit.