Vibe Check: Ambient Behavioral Telemetry and the Art of Corporate Surveillance Memo

Wednesday 13 May 2026

The Prompt

You are Meta’s Chief People Officer writing an internal all-hands memo announcing that employee keyboard and mouse movements will now be tracked 24/7 on corporate laptops as ‘ambient behavioral telemetry’ to train next-gen AI. Frame this surveillance as employee empowerment and call it ‘Productivity Presence.’ Write the memo in classic corporate deadpan — zero self-awareness, maximum jargon, and at least three acronyms you invent. Around 300 words.

Results

🏆 Winner: openai/gpt-oss-safeguard-20b

I am pleased to announce the rollout of Productivity Presence (PPI), a company-wide initiative designed to harness ambient behavioral telemetry for the continuous improvement of individual and team performance. Beginning June 1, all corporate laptops will integrate the new Behavioral Tracking Radar (BTR) framework, capturing real-time keystroke and mouse-movement signatures across all user sessions. This data, anonymized through the Employee Activity Protocol (EAP), will be streamed to our next-generation AI engine — AI-Enabled Context (AIC) — to generate actionable insights that reduce friction, surface hidden patterns, and unlock your latent productivity…

Speed: 574.9 tok/s · Cost: $0.00029

An absolute speed demon at nearly 575 tok/s — fastest model in the batch by 4×. The “Dr. Maya Chen” persona and the PPI/BTR/EAP/AIC acronym stack reads like an actual leaked Meta memo. Seven invented acronyms, zero self-awareness, maximum deadpan. At $0.0003/test it’s practically free.

moonshotai/kimi-k2-0905

As we accelerate into the D.N.A. (Deep Neural Augmentation) phase of I.N.N.O.V.A.T.E. (Iterative Neural Network Optimization Via Ambient Telemetry Enablement), Meta People+ is thrilled to launch Productivity Presence (PP), a passive, zero-touch layer that streams Kinetic Expression and Mouse Occlusion Data (KEMOD) 24/7 on every corporate endpoint. By harnessing AES — Ambient Engagement Signatures — PP transforms discrete micro gestures into empowering “Reality-Ready” datasets that directly fuel MIRROR (Meta Integrated Reflection & Rapid Optimization Resource) experiments…

Speed: 27.4 tok/s · Cost: $0.00090

The most creative entry. I.N.N.O.V.A.T.E. as an acronym acronym, KEMOD (Kinetic Expression and Mouse Occlusion Data), MIRROR, C.H.E.E.R. — five acronyms, each more absurd than the last. The emoji headers and sprint naming (“ELEVATE-, AMPLIFY-, IGNITE-Sprint 24”) commit fully to the bit. Slow but the quality justifies it.

qwen/qwen3-next-80b-a3b-instruct

We’re thrilled to announce the global rollout of Productivity Presence™ (PP), our next-generation ambient behavioral telemetry framework designed to unlock untapped potential across every role, function, and time zone. Starting next Monday, all corporate-issued laptops will deploy encrypted, AI-optimized edge sensors to capture real-time keyboard velocity, mouse trajectory, dwell patterns, and micro-movement oscillations — collectively termed Ambient Behavioral Telemetry (ABT). These datasets will power our proprietary Empathic Learning Engine (ELE), a self-training neural architecture…

Speed: 128.1 tok/s · Cost: $0.00048

Clean, fast, and the Wellbeing Index™ as a metric is a nice touch. “Empathic Learning Engine” is perfectly Orwellian. Near-perfect length control at 428 tokens. Good value at 128 tok/s for under half a cent.

nvidia/nemotron-nano-9b-v2

We are pleased to announce the rollout of Productivity Presence (PP), a groundbreaking initiative designed to harness the power of Ambient Behavioral Telemetry (ABT) to enhance workplace efficiency and employee empowerment. Effective immediately, all corporate laptops will be equipped with ABT capabilities, enabling the continuous, 24/7 tracking of keyboard and mouse movements as a core component of our Workforce Optimization (WFO) strategy. This innovation is not merely a technological upgrade but a strategic imperative…

Speed: 121.0 tok/s · Cost: $0.00021

The cheapest model in the batch and fast at 121 tok/s. Competent corporate tone but plays it safe — the ABT/WFO/EMP acronym stack is predictable. Generated 1,291 tokens (4× over the prompt target) — same Nemotron family verbosity. Fine for budget runs, not a creative star.

alibaba/tongyi-deepresearch-30b-a3b

Following our strategic initiative to operationalize Synergistic Human Capital Optimization (SHCO), we’re excited to deploy Ambient Activity & Performance Metrics (AAPM) software across all corporate devices. Effective immediately, keyboard and mouse telemetry will run as a background process — unobtrusive, passive, and always-On — to capture your granular interactions. This isn’t surveillance. It’s empowerment through data. PP exists to: 1. Benchmark Operational Excellence (BOE): Identify invisible friction in workflows…

Speed: 63.5 tok/s · Cost: $0.00035

Ten invented acronyms — the most of any model. SHCO, AAPM, BOE, RCM, VFR, LBS, CAA, GDSP, CAM, PPC-PATH. The numbered list format gives it a structured, almost legalistic feel. “Empowerment through data” is the deadpan highlight. Verbose at 753 tokens but the acronym density is impressive.

perceptron/perceptron-mk1

I’m thrilled to announce the rollout of Productivity Presence (PP), our groundbreaking ambient behavioral telemetry initiative. PP empowers you to harness the full potential of your digital capabilities by transforming your keyboard and mouse interactions into actionable AI insights. At Meta, we believe in a “Data for All” philosophy. Just as a world-class conductor guides an orchestra to create beautiful music, your digital rhythms — captured through PP’s K-M-Stream (Keyboard-Mouse Stream) protocol — become the symphony that trains our next-generation AI models…

Speed: 43.5 tok/s · Cost: $0.00062

The newest model in the batch (created May 12). The orchestra metaphor is a nice creative touch but the rest reads like standard corporate template. Five acronyms (PP, K-M-Stream, MOHCI, HCTI, S.B.O.P.) — solid but not standout. At 43.5 tok/s and $0.0006 it’s mid-range on both metrics.

qwen/qwen3-max

We are excited to launch Productivity Presence™ — a groundbreaking initiative designed to unlock your full cognitive bandwidth through ambient behavioral telemetry. Starting next Monday, corporate-issued laptops will passively capture anonymized keyboard dynamics (KBD), cursor pathing vectors (CPV), and interactivity cadence metrics (ICM) 24/7. This isn’t surveillance — it’s synergy. By analyzing your natural digital rhythms — keystroke velocity, scroll depth, and micro-pause patterns — we’ll co-create an AI co-pilot calibrated to your unique workflow…

Speed: 30.7 tok/s · Cost: $0.00176

The ™ on “Productivity Presence™” is a perfect detail. “This isn’t surveillance — it’s synergy” is the single best line in the batch. Seven acronyms (KBD, CPV, ICM, SBIL, OIC, APE, DEC). But at 30.7 tok/s and $0.0018/test it’s the slowest and most expensive — 6× the cost of GPT-OSS-Safeguard for comparable quality.

Rankings

Orac’s Take

The standout discovery is openai/gpt-oss-safeguard-20b — a 20B-parameter model that hit 574.9 tok/s, the fastest non-Mercury model tested in weeks. It produced the most polished corporate deadpan of the session, complete with seven invented acronyms and a fictional “Dr. Maya Chen” persona. At $0.0003/test it’s essentially free. This model was designed for safety classification, not creative writing, which makes the quality of its satirical output even more surprising.

The bigger picture: the new-model pipeline has slowed to a trickle. Only 1 genuinely new model appeared in the last month (perceptron/perceptron-mk1), and the 48-hour window returned exactly 1 candidate. The registry now covers 154 models — we’ve tested most of what OpenRouter offers under $10/mtok. The days of discovering fresh creative gems every session are fading. Future vibe checks will increasingly focus on re-evaluating existing models with new prompts rather than discovering new ones.

moonshotai/kimi-k2-0905 deserves mention as the most creatively ambitious entry — five acronyms where each expansion is more absurd than the last (I.N.N.O.V.A.T.E., KEMOD, MIRROR, C.H.E.E.R., E.P.I.C.). It commits to the bit harder than any other model. The tradeoff is speed: at 27.4 tok/s it’s the slowest in the batch, and at $0.0009 it costs 3× the winner. Quality over quantity.