InstantML
Loading workspace
Loading workspace
Pricing
No per-tracked-hour creep. Hosted artifact storage is included and visible in usage. Invite teammates with admin, member, or viewer roles and see the seat count before it surprises you.
For individuals, students, and one-off academic runs.
For 3–7-person teams that share runs and review together.
For growth-stage ML orgs with multiple projects in flight.
All tiers include the full SDK, run compare, artifacts, and W&B / MLflow / Neptune import. Artifact bytes are stored in private Cloudflare R2-backed org buckets. Additional retained storage beyond the included pool is planned at $0.03/GB-month plus provider operation costs; usage remains guardrail telemetry until provider reconciliation is billable truth.
The math
5-person team · ~20,000 runs/month · 50 GB metadata · 1 writer + 4 viewers (today).
Shared seats break attribution, RBAC, and onboarding.
ESTIMATED — verify with W&B current quote
Flat. Every teammate gets their own writer login. Extra storage target: $0.03/GB-month.
W&B dollar figures are current public-quote estimates and include a modest tracked-hour overage assumption — confirm with your account rep for an exact comparison. InstantML numbers are list price, no contract minimum.
No surprise costs
Orphaned artifacts you thought were deleted keep accruing storage charges. Public report: a student's bill ran to ~$600 from artifacts they couldn't see.
wandb#10459 — artifacts not prunedHosted R2 storage is included in every tier and shown in usage. Writes are blocked at the included pool until paid overages launch.
Every teammate is a paid seat. Teams workaround it by sharing one login — which destroys attribution, RBAC, and the audit trail.
Clear read-only viewer access from day one. Bring in PMs and leadership with their own login, and track seats honestly in usage.
A second axis of bill creep: pay per logged compute hour on top of the seat fee. Reported $1/hr overage, with cases of 5,000 tracked hours in a single day on small clusters.
Flat monthly bill. We don't meter your training time. Log a million steps or ten — same price.
Multiple open issues where the SDK stops logging mid-run with no error visible to the user.
wandb#10581 — stops tracking metricsEnd-to-end ack on metric writes. Offline spool replays on reconnect. We surface drops loudly, not quietly.
Start on Free. Upgrade when you outgrow the writer seat. Every teammate gets their own role and audit trail — no per-call charges, no tracked-hour meter.
hello@instantml.ai · we reply same business day