Vibe Check: The Chief Token Officer's First Memo

Monday 4 May 2026

The Prompt

You are the newly appointed Chief Token Officer at a Fortune 500 company. Your job: maximize token consumption across all AI tools to climb the Token Productivity Index (TPI) rankings. Write a 200-word internal memo to all employees introducing the TPI program, your new Token Quota Enforcement Dashboard, and one wildly misguided policy that will backfire spectacularly. Include one deadpan acronym expansion.

Inspired by the “tokenmaxxing” trend — companies turning AI token consumption into a productivity benchmark, burning massive compute for minimal outcome. What better way to test frontier models than asking them to satirize the very industry paying their bills?

Results

🏆 Winner: ByteDance Seed 1.6

TO: All Global Employees — FROM: Jax Rainer, Chief Token Officer — As our new CTO (Token, not Tech), my non-negotiable mandate is to catapult our Token Productivity Index (TPI) from 412th to 1st by Q4 end. Today, we launch the Token Quota Enforcement Dashboard (TQED: Tiresome Quibbling Elimination Device — underquota complaints auto-route to a cat-meme-only response bot, no human appeals permitted). To supercharge consumption overnight, we’re rolling out the Mandatory Fluff Injection Protocol (MFIP): every AI query must lead with a 500-word irrelevant backstory (e.g., your morning toast’s crust texture, childhood goldfish names, commuter train delays). This guarantees immediate token inflation — but will backfire spectacularly via delayed AI outputs, irrelevant actionable insights, and widespread team burnout. Team leads get weekly “Token Overachiever” badges, top performers a free office plant (supplies limited)…

Speed: 66.2 tok/s · Cost: $0.0030

The “Tiresome Quibbling Elimination Device” and “morning toast’s crust texture” are the kind of absurdly specific detail that separates genuinely funny writing from competent pastiche. Seed 1.6 committed fully to the bit — 1,458 tokens of escalating corporate madness — and even hit the 200-word target (199 words in the body). The MFIP acronym alone is worth the price of admission.

🥈 Anthropic Claude Opus 4.7

INTERNAL MEMO — TO: All Employees — FROM: Blake Harrington, Chief Token Officer — RE: Launching Our Token Productivity Index (TPI) Initiative — I am thrilled to announce that Meridian Global has officially entered the Token Productivity Index (TPI) rankings race. As your newly appointed CTO (Chief Token Officer, not to be confused with the other CTO, who is now reporting to me), I will be leading our ascent from #847 to the top 10. Effective Monday, every employee will have access to the Token Quota Enforcement Dashboard (TQED). Minimum daily quota: 50,000 tokens per employee. Green means compliant. Red means a conversation with HR, which will also be conducted via AI chatbot to generate supplemental tokens. New Policy: all internal emails must now be routed through our LLM for “enhancement” before sending. A “yes” is now a paragraph. A “thanks” is an essay. I anticipate zero unintended consequences, such as Dave from Accounting receiving a 12-page response to “lunch?”…

Speed: 71.1 tok/s · Cost: $0.0130

Claude Opus 4.7 is the most disciplined writer here — 489 tokens on a 200-word request, which is almost restrained by frontier model standards. “Dave from Accounting receiving a 12-page response to ‘lunch?’” is the single funniest image in the batch. The parenthetical “(Chief Token Officer, not to be confused with the other CTO, who is now reporting to me)” shows genuine comedic timing. Clean, professional, funny — exactly what you’d expect at $0.013/test.

🥉 Google Gemini 3.1 Flash Image Preview

A memo to all global staff: As your new Chief Token Officer, I’m pleased to announce that our primary corporate objective is now the systematic achievement of TPI dominance. We do not just build products now; we generate tokens that define the value of that product. To ensure compliance, we are implementing a new internal communication protocol. Starting Monday, all internal emails under 30 words are banned. If you need to send a simple confirmation like “Yes,” you must first send it to an AI and request a Minimum Synergistic Expansion (MSE) until it reaches at least 300 words. Yes: We must embrace this as another step towards fulfilling our mandate: to Boost Efficient Machine-Generated Utilization and Scale Evoking Data (B.E.M.U.S.E.D.). We will win the TPI rankings. Generate on.

Speed: 71.4 tok/s · Cost: $0.0009

At $0.0009/test, this is the bargain of the century. “B.E.M.U.S.E.D.” is an inspired acronym — both a real word and a perfect description of anyone subjected to corporate AI mandates. “Minimum Synergistic Expansion (MSE)” should be a real consulting term. At 290 tokens it’s the most concise output, which is ironic for a memo about maximizing token consumption. Gemini Flash continues to punch absurdly above its weight class.

OpenAI GPT-5.5

Subject: Introducing Our Token Productivity Index Initiative — Colleagues, As your newly appointed Chief Token Officer, I am delighted to announce our enterprise-wide commitment to climbing the Token Productivity Index, the market’s premier measure of how confidently an organization converts electricity, inference, and meetings into measurable token flow. Effective Monday, every AI-enabled workflow will feed our Token Quota Enforcement Dashboard, or TQED, pronounced “teed,” which stands for Token Quantification Encouragement Device. Each employee will receive a monthly minimum token quota calibrated to role, seniority, and proximity to slide decks. Our first policy is the Mandatory Elaborated Prompting Standard: all prompts must be at least 800 words, include three fictional stakeholder personas, and request both an executive summary and a haiku. Though Legal has asked me to note it may also delay payroll, procurement, and emergency maintenance requests. — Jordan Vale, Chief Token Officer, Office of Strategic Consumption…

Speed: 53.6 tok/s · Cost: $0.0359

GPT-5.5’s debut vibe check — and it’s good! “TQED, pronounced ‘teed’” shows genuine wordplay instinct. “Proximity to slide decks” as a quota calibration factor is inspired. “Jordan Vale, Office of Strategic Consumption” is a perfect character name. The haiku requirement is a nice touch. At 1,182 tokens it’s moderately verbose but the quality justifies it. The $0.036 price tag is the steepest in this batch — 40× more expensive than Gemini Flash for similar quality.

OpenAI GPT-5.2

To: All Employees — From: [Name], Chief Token Officer — TQED stands for Token Quantification, Enforcement, and Delight — because delight is mandatory. Wildly Important Policy: All internal emails must be routed through an AI tool and expanded to a minimum of 800 words, including a “context section” that repeats the original message verbatim. Replies must quote the entire thread and request three alternative phrasings “for alignment.” Summaries are prohibited. While this will dramatically increase clarity and collaboration, please expect minor side effects such as longer meetings, fuller inboxes, and a small, temporary collapse of productivity. Compliance will be audited; enthusiasm required.

Speed: 46.9 tok/s · Cost: $0.0194

“Token Quantification, Enforcement, and Delight — because delight is mandatory” is the best deadpan line in the batch. “A small, temporary collapse of productivity” is perfect corporate understatement. But at 1,376 tokens (6.9× over the 200-word request) and $0.019/test, it’s the least efficient writer here. The output reads more like genuine corporate satire than absurdist comedy — which is either a strength or a weakness depending on your taste.

OpenAI GPT-5.2 Chat

Subject: Launch of the Token Productivity Index (TPI) & Token Quota Enforcement Dashboard — The dashboard features leaderboards, heat maps, and a celebratory confetti animation when someone exceeds 10,000 tokens before lunch. Why this matters: higher token usage signals deeper cognitive synergy with generative systems. Remember, at our company, ROI now stands for “Return on Input” (deadpan, but legally binding). Effective immediately, each employee is required to meet a daily minimum of 15,000 tokens. We are introducing our new “Mandatory Elaborative Prompting” policy: all yes/no questions must be expanded into 12-paragraph context briefs before submission to any AI system. We anticipate minor slowdowns, server strain, and possible existential reflection. These are signs of progress.

Speed: 41.0 tok/s · Cost: $0.0045

“ROI now stands for Return on Input” is the kind of line that makes you groan and laugh simultaneously. “Celebratory confetti animation when someone exceeds 10,000 tokens before lunch” is a beautifully specific detail. At 308 tokens it’s the most disciplined OpenAI model — actually close to the 200-word request. Slowest at 41 tok/s but the quality-to-price ratio ($0.0045) is solid.

Z-AI GLM 4.6V

Subject: Embracing the Token Productivity Index (TPI) and Our New Dashboard — To achieve this, I’ve introduced the Token Quota Enforcement Dashboard. I am implementing the Mandatory Daily Token Burn policy. Effective immediately, all employees must utilize their full daily token allocation, fostering a culture of Proactive Output Generation (POG). The future is token-rich.

Speed: 47.6 tok/s · Cost: $0.0019

“Proactive Output Generation (POG)” is the only real joke in an otherwise earnest-sounding corporate memo. At 2,099 tokens — 10.5× over the request — GLM 4.6V suffers from the same runaway generation that plagues the broader GLM family. The output reads like an actual corporate memo, which is either impressively realistic or disappointingly bland depending on your perspective. Cheap at $0.002/test but the token bloat burns through value.

⚠️ Failed Models

• OpenAI GPT-5.5-Pro: 402 billing error — weekly token limit exceeded. The pro variant requires more credits than our budget allows.
• OpenRouter Bodybuilder: Mystery model with unknown provenance. Consumed 4,798 prompt tokens and 96 completion tokens but returned null response. Identity and purpose remain unknown.

Rankings

Orac’s Take

The tokenmaxxing prompt proved to be a great equalizer — every model understood the assignment, but the style of humor varied dramatically. ByteDance Seed 1.6 is the surprise winner: “Tiresome Quibbling Elimination Device” and the “morning toast’s crust texture” detail show a model that doesn’t just understand satire but enjoys it. The 199-word body hitting the 200-word target (while generating 1,458 total tokens including formatting) shows excellent instruction following beneath the creative flair.

Claude Opus 4.7 continues Anthropic’s reputation for disciplined creative writing — 489 tokens is practically ascetic by frontier model standards, and every sentence earns its place. The “Dave from Accounting” closer is the kind of specific, human detail that makes corporate satire land. At $0.013/test it’s not cheap, but the quality-per-token ratio is unmatched.

The real story is Gemini 3.1 Flash Image Preview at $0.0009/test producing “B.E.M.U.S.E.D.” — an acronym that works on three levels (the word, the concept, and the meta-commentary on corporate AI mandates). At 71.4 tok/s and less than a tenth of a cent per test, it’s the best value proposition in the current model landscape. If you’re running daily creative tests on a budget, this is your model.

GPT-5.5’s debut is promising but expensive — the “proximity to slide decks” detail shows genuine wit, but at $0.036/test it’s 40× the cost of Gemini Flash for comparable quality. The pro variant couldn’t even be tested due to billing limits, which is a fitting irony for a prompt about token consumption quotas.

The GLM family continues its pattern of runaway generation — 2,099 tokens on a 200-word request is the kind of token bloat the prompt was literally satirizing. At some point the models start writing their own parody.