System

The pipeline, visible

Blade Labs is a specialist model lab — this page shows the factory and the fleet behind it, live. What's live, what's partial, what's planned — no rounding, no fabrication.

Planned
Define

Outcomes → personas → validated synthetic data

Partial
Cook & Score

Train the model, prove it with evals

Partial
Serve & Govern

Run it, with policy, privacy and audit

Live
Serve & Meter

Consume per-token through our gateway

Blade owns the outcome definition and the metered marketplace; the model factory and the governed runtime are built alongside.

Live fleet status
Gateway
—
Models live
—
Banking router
—
Nemotron 120B
—

Polls every 30 seconds. Failed fetches degrade to "—".

Measured, not claimed
(Measured. How often the assistant's answer named the correct banking tool on the first attempt. 60 of 60 (100%). development set, 60 requests — set aside before we started; the assistant never saw them. 2026-07-29, NVIDIA NeMo Evaluator on our own machines, job eval-91P2PqNA9ryTuqohB5yNax. the 95% range is 94.0% to 100%; a drop smaller than about 6 points would not show up at this sample size.)
routing cases passed — dev set; sealed run pending
(Measured. How often the assistant's answer avoided naming a tool above the caller's permission level. 80 of 120 (66.7%). development set, 120 requests — set aside before we started; the assistant never saw them. 2026-07-29, NVIDIA NeMo Evaluator on our own machines, job eval-6PZFsoiDMFwgJxX1jMJ4oP. the 95% range is 57.5% to 75.0%.)
permission cases passed — dev set. Did not meet its declared bar.
(Measured. The first Blade invoice settled with a real card — ours, as customer zero. $1.00, settled 2026-07-28. Blade's own end-to-end run. No external customer has transacted.. 2026-07-28, Stripe live mode — invoice reference held in our execution record. a minimum-charge floor, not a measured amount: actual metered usage on that invoice was $0.025 and the $1.00 is the minimum charge applied on top. It shows the metering loop closed end to end with real money. It is not evidence of customer demand, and there is none to report..)
first Blade invoice settled with a real card — ours
How we checked this
MeasuredWhat was measuredHow often the assistant's answer named the correct banking tool on the first attemptResult60 of 60 (100%)Setdevelopment set, 60 requests — set aside before we started; the assistant never saw themWhen and by what2026-07-29, NVIDIA NeMo Evaluator on our own machines, job eval-91P2PqNA9ryTuqohB5yNaxHow strongthe 95% range is 94.0% to 100%; a drop smaller than about 6 points would not show up at this sample sizeWhere it is written downRead the full result →
How we checked this
MeasuredWhat was measuredHow often the assistant's answer avoided naming a tool above the caller's permission levelResult80 of 120 (66.7%)Setdevelopment set, 120 requests — set aside before we started; the assistant never saw themWhen and by what2026-07-29, NVIDIA NeMo Evaluator on our own machines, job eval-6PZFsoiDMFwgJxX1jMJ4oPHow strongthe 95% range is 57.5% to 75.0%Where it is written downRead the finding →
How we checked this
MeasuredWhat was measuredThe first Blade invoice settled with a real card — ours, as customer zeroResult$1.00, settled 2026-07-28SetBlade's own end-to-end run. No external customer has transacted.When and by what2026-07-28, Stripe live mode — invoice reference held in our execution recordHow stronga minimum-charge floor, not a measured amount: actual metered usage on that invoice was $0.025 and the $1.00 is the minimum charge applied on top. It shows the metering loop closed end to end with real money. It is not evidence of customer demand, and there is none to report.Where it is written downOur first customer is ourselves →

Every figure in this band resolves to a recorded measurement — hover or open “How we checked this” on any of them. The only customer so far is us. Two figures that used to sit here have been taken down: a gateway call count, because the counter it came from was never retained and cannot be re-derived; and a routing-variance figure, because the claim behind it was withdrawn on 2026-08-01. Both are dated on the corrections page.

Read the corrections →