Pay for GPU time, nothing else
No seats, no subscription, no annual commitment. You top up a credit balance, and each design run draws down the GPU time it actually used. Every run is quoted before it starts.
Estimate a run
Cost scales with the total residue count of your complex (target + binder) and the tool you pick. Drag the sliders to price a run:
Estimated from 85%-of-variance production fits. The studio re-quotes this exact number before you submit, and you are charged actual GPU time — runs that finish early cost less.
How billing works
- Top up a balanceAdd credits in any amount from $5 up. Card payments are handled by Stripe; SeqDesign never sees your card details.
- See the quote before you submitThe studio prices your run — per design and in total — as soon as you have picked a target and a tool. If your balance is short, it tells you by how much before anything is queued.
- Pay for what actually ranCharges follow real GPU time. Runs that converge early cost less than the estimate. Failed jobs are not billed.
What's included at no extra cost
The studio
Structure search, the 3D viewer, hotspot picking, chain trimming, sequence tools and results browsing are free. You only pay when you dispatch a job to a GPU.
Storage and results
Designs, structures, metrics and CSV exports for your experiments are kept and served at no additional charge.
Target MSAs
Multiple-sequence alignments are computed once per target and cached, so a 50-design campaign against one target doesn't pay for 50 alignments.
Where the numbers come from
The estimate is not a list price we picked — it is a regression fit against production job history, refreshed when the fleet or the tools change. Cost per design is flat near a per-tool floor up to a knee, then rises super-linearly as GPU memory and attention pressure scale with residue count.
| Tool | Type | Max residues | Fit quality |
|---|---|---|---|
| BindCraft | Binder design | 500 | r² 0.85 |
| BoltzGen | Binder design | 1000 | r² 0.88 |
| ProteinHunter | Binder design | 550 | r² 0.76 |
| Boltz-2 | Structure prediction | 1000 | r² 0.93 |
| ESMFold2 | Structure prediction | 1000 | Quoted at submit |
| Protenix | Structure prediction | 1000 | Quoted at submit |
Jobs above a tool's residue ceiling are rejected at submission rather than queued and failed, so you are never charged for a run that could not have completed.
Would you rather we ran it?
For campaigns that need experimental design, target triage or wet-lab handoff, our team takes the project end to end. Pricing there is per project, not per GPU hour.