Open models like ESMFold, AlphaFold, Boltz and ProteinMPNN are powerful and free, but they return a number, not a judgment — which model fits the job, whether the prediction is reliable, and what to do when two models disagree. Orbion adds that layer.
You're a technical team already using open models for research, wondering whether that's enough to make production decisions.
The Difference
What Actually Separates Them
Open protein models are one of the best things to happen to the field. ESMFold, AlphaFold, Boltz and ProteinMPNN are strong, improving fast, and free. For research and exploration, running them directly is the right move.
The gap opens when a decision rides on the answer. An open model gives you a structure or a score. It does not tell you which model to reach for on this target, whether this particular prediction is reliable, or what to conclude when two models disagree. It also does not touch the things a wet-lab team cares about most — expression yield and developability.
That interpretation layer is the work. Orbion runs the same open models, plus its own, and wraps them in the judgment that turns a raw number into a call you can act on.
At A Glance
The Model Is The Small Part
Side By Side
How They Compare
| Open Models (DIY) | Orbion | |
|---|---|---|
| What You Get Out | A Number Or A Structure | A Ranked, Flagged Decision |
| Model Selection | You Decide | Routed For You |
| When Two Models Disagree | You Reconcile It | Reconciled And Flagged |
| Confidence Signal | Raw, Uncalibrated | Calibrated, Gated |
| Expression Yield & Developability | Not Covered | Predicted From Sequence |
| Membrane & Hard Targets | Varies By Model | A Validated Strength |
| Cost | Free (Plus Your Time) | Subscription |
| Best For | Research & One-Offs | Decisions With A Downside |
The Hidden Work
What You End Up Owning
A Number, Not A Verdict
The model returns a value. Deciding what it means for your target is the part it leaves to you.
You Pick The Model
Which of the open models suits this protein, and this question, is a research call you make every time.
No Confidence Flag
Raw model scores aren't calibrated success rates. A confident-looking number can be the unreliable one.
You Keep Up With The Field
New open models ship constantly. Tracking, testing and adopting them is an ongoing job.
When Open Models On Their Own Are The Right Call
For research, exploration and one-off analyses, running open weights yourself is excellent, flexible and free. Orbion earns its place when a decision — and its downside — rides on the answer, and you need the output to be calibrated and flagged, not raw.
Where Orbion Fits
The Judgment Layer, Prebuilt
- Orbion runs the same open models, plus its own, and adds the layer that makes them decision-grade: routing, multi-model reconciliation, calibrated confidence, and low-confidence flagging.
- It also covers what open structure models don't — expression yield, developability and stabilizing mutations — so the output maps to what you do at the bench.
The model is the cheap part. The product is the judgment around it — model routing, calibrated confidence, low-confidence flags, and a workflow that stops a bad prediction from becoming a bad decision.
The Evidence
Backed By Bench Results, Not Just Claims
Prospective and blind where noted. The signal is in the ranking — absolute numbers still need your wet lab, which is why every engagement starts with a blind benchmark on your own targets.
Go Deeper
Related On Orbion
FAQ
Common Questions
Orbion routes between leading open structure and sequence models and combines them with its own, choosing per target and per question. We report results, not model internals.