Model FactoryTurn your own cases into models that keep improving.
A model that keeps improving
Train on your own cases, check the result against the base model, serve the winner, and let its production runs become the next round's cases.
- 01Your casesBenchmarks from production work, with a checker that scores each attempt 0 to 1.
- 02TrainReinforcement learning on an open model, with Cura 1T as the teacher.
- 03CompareHeld-out scores against the base model, task by task, then side by side.
- 04DeployServe the winner under one stable name your agents call.
Production runs come back as the next round's cases, so the model keeps improving.
See what changed
Every run explains itself: the score to trust, the checkpoint to try, and which tasks moved.
- Against the base modelHeld-out reward at every evaluation, drawn against where the base model started.
- Task by taskSee which held-out tasks improved, regressed or held, and open any attempt to read it.
- Overfitting flaggedWhen the held-out score falls while training keeps rising, the run says so.
Recommended checkpointStep 40Try vs baseDeploy
Base model0.39Before training
Best held-out0.61Step 40
Gain from training+0.22Best minus base
What changed, task by task
Held-out tasks, base model against step 40: 31 improved, 6 regressed, 63 unchanged.
- Retro-authorization after admission
0.00 → 1.00 - Out-of-network imaging request
0.00 → 1.00 - Referral missing the physician's order
0.50 → 1.00 - Duplicate submission
1.00 → 0.00
+ 96 more held-out tasksOpen each attempt
Compare in the Playground
Ask the base model and a checkpoint the same question, side by side, before anything ships.
- Side by sideBoth models answer the same prompt at once.
- Held-out tasks to tryStart from tasks the model never trained on.
- Nothing savedThe Playground is for looking. Nothing asked there is kept.
Assistant AQwen3.8 27B · base model
Assistant BStep 40 checkpoint
Referral for lumbar MRI, no conservative therapy documented. Recommend?
Approve: imaging is medically necessary for back pain.
Pend: criteria need six weeks of conservative therapy first. Request the PT notes.
Try a held-out task
case-204Retro-authorization requested two days after admission…case-311Out-of-network MRI, in-network imaging center 12 miles away…case-087Duplicate request for a CT already approved last week…
Ask both models something…
Both answer at once. Nothing here is saved.
Deploy what wins
Serve the checkpoint you chose under a stable name your agents and API keys call.
- Stable API nameAn OpenAI-compatible model name that stays the same from version to version.
- New versions, one nameMake another checkpoint live, or roll back, from the deployment.
- Inside your boundaryTraining and serving stay within your compliance boundary.
Deployments
Served to this organization's API keys
prior-auth-specialistServing
- v2Step 40 · 0.61 held-outLive
- v1Step 20 · 0.53 held-outRestore
"model": "prior-auth-specialist"
appeals-reviewerv1 · step 30 · 0.57 held-outServingintake-triagev3 · step 50 · 0.72 held-outServing
Requests today12,480
Median latency1.4 s
Versions served2
Two ways to work
Have our forward-deployed researchers train it with your experts, or run Model Factory yourselves. Either way, the model is yours.
Forward deployedOur team builds it with you
Our forward-deployed researchers work beside your experts, write the benchmark from your work, and hand back a model that beats the base model on it.
- Researchers on your workflow
- Benchmarks written on your own outcomes
- The weights are yours
Self-serveYour team runs it
Your team launches runs in Model Factory, reads every result, and deploys the checkpoint it chooses.
- New Training Run in five steps
- Playground and task-by-task results
- Deploy and roll back from one place
Build your first model with us
Bring one workflow your agents run every day. We'll show you what a model trained on it can do.