The public AI Benchmarks page shows how models perform on real MyTabulon agent work. RPB means Real Performance Benchmark: an evidence-adjusted score built from task completion, tool accuracy, reasoning, safety, user sentiment, latency, and cost efficiency.
The public AI Benchmarks page shows how models perform on real MyTabulon agent work. RPB means Real Performance Benchmark: an evidence-adjusted score built from task completion, tool accuracy, reasoning, safety, user sentiment, latency, and cost efficiency.
RPBReal Performance Benchmark
ScoreEvidence-adjusted public model score
EvidenceProduction signals, tool ledger, feedback, golden tasks
PrivacyNo prompts, payloads, customer data, or record IDs
Public URLs/benchmarks, /benchmarks/blog, /benchmarks/data, /benchmarks/methodology, /benchmarks/run, /benchmarks/faq
Best read withConfidence, sample size, domains, traces, latency
Open /benchmarks for the public leaderboard, realtime and 24-hour views, model cards, search, period filters, domain winners, sanitized traces, and golden-task evidence.
RPB is not a lab-only exam or a model popularity score; it measures real agent behavior while doing business work inside MyTabulon and avoids perfect-score claims from small samples.
Use /benchmarks/methodology to explain score weights, confidence, privacy boundaries, anti-gaming controls, and why small samples show as warming or insufficient data.
Use /benchmarks/blog for the research note, /benchmarks/data for JSON endpoints and schema, and /benchmarks/run for evaluation setup.
Use /benchmarks/faq when a customer, partner, or AI company needs a plain-language explanation of RPB, tool call accuracy, privacy, and confidence.
Public evidence may show model label, tool names, domain, outcome, broad signals, and sample size; it does not expose prompts, customer names, record IDs, files, or raw payloads.
Domain winners help explain where a model is strongest, such as CRM, accounting, documents, operations, inventory, memory, integrations, research, or reasoning.
Open Business Settings, complete the company profile, set the currency, upload the logo, confirm billing, then add clients, inventory, or leads depending on what the business sells.
The app checks the active workspace's effective subscription, subscription state, and usage caps before opening or changing protected records. Owner-created workspaces can inherit the owner's account plan from another billing-home workspace, while joined team workspaces use that company's owner plan. A 402 response usually means the effective plan is free, inactive, or has reached a limit for that feature. File Manager checks file count, storage, and per-file upload size: Free 50 MB, Plus 250 MB, Pro 2 GB, and Max 10 GB. AI Images checks image credits, daily and personal caps, model access, source limits, File Manager count, and storage. Maximo Cloud Computer checks computer minutes, daily/monthly caps, max run length, and concurrency. Gemini Live starts on Pro, while Grok Voice remains available on paid plans. Support also follows the plan: Free uses guide, community, and email; Pro and Max include monthly Google Meet allowances.
RPB means Real Performance Benchmark. It is MyTabulon's evidence-adjusted public model score for real agentic business work, combining task completion, tool call accuracy, reasoning, safety, user sentiment, latency, and cost efficiency.
Start with RPB, then check confidence, sample size, tool accuracy, median tool latency, domain winners, and sanitized traces. A high RPB with low confidence is promising but still collecting evidence. A model can also lead in one domain, such as CRM or accounting, without being the overall leader.
No. Public benchmark evidence can show model labels, domains, tool names, statuses, broad outcomes, scores, and sample counts. It does not show private prompts, raw payloads, business names, customer names, record IDs, file contents, or workspace-only data.
The benchmark is designed to be hard to game because it uses production tool outcomes, anonymized action evidence, user feedback, hidden or rotating golden tasks, confidence thresholds, and versioned scoring. It does not accept a model's own claim that it completed a task.
Global search can find feature pages plus clients, leads, inventory, invoices, quotes, deals, projects, tasks, files, expenses, team members, and AI chats within the active company when the signed-in user's role allows it. Member file results are limited to files and folders they uploaded, added, or generated, and sensitive owner/admin destinations stay hidden.
Use Ctrl K or Meta K for the command palette, ? or Ctrl / for the reference, Ctrl B or Meta B for the sidebar, g chords for navigation, n chords for quick create, and r r to refresh page data.
Payment links use long random tokens. Share the generated link, receipt, PDF, or email with customers, not a dashboard URL. The invoice number helps the business identify a document but is not the secret.
Business Memory stores durable company knowledge for Maximo AI: facts, preferences, style, source records, private notes, approved memories, retention rules, provenance, and conflict checks. It is available only to business owners and administrators. Company memories describe shared business facts; first-person or personal work memories stay private to the originating user. Members, Viewers, and Employee Viewers cannot open memory, search it through Maximo chat or voice, receive selected memory in AI context, or have their chats silently captured into company memory.
Who can add or manage Bring Your Own Model (BYOM) connections?
Only Business Owners and Administrators have permissions to manage BYOM connections and import/edit models under Settings > Integrations > Bring Your Own Model. Workspace Members and Viewers cannot manage connections.
Are BYOM models available to everyone in the workspace?
Yes, once a BYOM model is configured and enabled by an owner or administrator, it is available company-wide to all workspace members (Owners, Admins, Members, Viewers) through the model selector in Maximo AI chat.
