The Gym Software The Check Demo For Labs Go X Gym Pitch Request a sample →
FOR AI LABS

The internet ran out.
Businesses didn’t.

Real companies produce the data models still can’t generate: messy inboxes, angry customers, regulatory filings, pricing wars. We turn that into RL environments you can train against.

28
Environments live today.
1.4M+
Scored tasks per environment, on average.
63
Anonymization categories, applied pre-transit.
8
Modalities: text, code, docs, calls, screen, POS, logs, forms.
WHAT WE SELL

Not screenshots. Not synthetic.
Real business, calibrated for training.

01

Real data, not synthetic proxies

Sourced from operating businesses, not simulated. Every task is one your team already saw, with the outcome the business actually lived with.

02

Multimodal from the start

Docs, communications, spreadsheets, screen recordings, phone calls, POS transactions, form filings. All from the same business, in the same clock.

03

Calibrated for the frontier

Every task is scored on your current best model. Environments graduate when they stop discriminating — we ship you harder ones.

MRW-BENCH · Q3 2026

The businesses your models can't run yet.

Composite score across scheduling, refunds, filings, hiring, pricing. 100 = matches human ops lead. Higher = better than the CEO.

# MODEL SCORE RUNS Δ 30D
01
Cypher 5.2 Ultra Cypher Labs
72.4 14,208 ▲ +4.1
02
Orion 6.1 Max Lab A
68.9 22,140 ▲ +2.7
03
Fable 5.1 Max Lab B
66.2 18,442 ▲ +1.9
04
Aurora Turbo Lab C
64.8 31,088 ▼ -0.3
05
Zenith 5.5 Lab A
61.3 27,600 ▲ +3.2
06
Aster 5 Fast Lab D
58.7 9,412 ▲ +5.1
07
Meridian 405B Lab E
55.4 6,209 ▼ -1.8
—
Human ops lead baseline · SF · 3yr experience
100.0 — —

LEADERBOARD FROZEN 2026-09-10 · SCORING METHODOLOGY UNDER NDA

CATALOG

Twenty-eight environments live today.

More every month. Each anchored on a real running business.

MOBILITY

Micro-fleet

150-unit urban shared mobility fleet. Dispatch, permits, refunds, pricing. 1.4M events / 6 mo.

HOSPITALITY

Nine-location café

Small chain. Payroll, POS, inventory, complaints. 890K events / 8 mo.

CLINICAL

Dermatology practice

Regional clinic. Intake, prior auth, coding. Fully HIPAA anonymized. 220K events / 12 mo.

LOGISTICS

Two-state trucking

18 rigs. Dispatch, DOT filings, weigh-stops. 610K events / 9 mo.

LEGAL

Immigration practice

Boutique firm. Filings, translations, USCIS. 340K events / 14 mo.

RETAIL

DTC apparel

$8M brand. Returns, ads, fulfillment. 1.1M events / 10 mo.

SERVICES

Plumbing dispatch

Regional shop. Field ops, quotes, callbacks. 280K events / 11 mo.

+21

More on request

Real estate, agencies, nonprofits, education, construction, utilities. Full list under NDA.

ENGAGEMENT

Three ways to work with us.

A

Environment access

Metered access to a live environment. Score submissions, compare model versions, export eval reports. Standard 12-month term.

B

Bespoke environments

Tell us the vertical or the failure mode. We source the business, build the gym, calibrate to your models. Six to ten weeks.

C

Full-stack dataset

Raw anonymized dataset (no environment). For SFT, evals, cold-start RL. Named-run exclusivity available.

Train on the real world.
Not the copy of it.

Request a sample environment. Score your best model against a real business. Then decide.