Vibe Check: Burning Dinos for the Singularity
21 April 2026 · 4 new models tested · Inspired by the Stargate energy buildout
The Prompt
Write a 200-word satirical “customer testimonial” from a fictional AI company CEO explaining why their new data center needs its own natural gas microgrid. The CEO should be cheerfully oblivious to the irony. Include at least one absurd corporate buzzword and one accidentally honest admission.
Results
🏆 Winner: qwen/qwen3.5-27b
At OmniMind AI, we believe the future is powered by passion—and methane. Installing our dedicated natural gas microgrid was the most forward-thinking decision we’ve made this quarter. While competitors worry about grid instability, we’ve achieved total energy sovereignty through what we call ‘combustion-aligned scalability.’
Critics ask about carbon footprints, but we view emissions as simply ‘visible progress.’ The microgrid ensures our Large Language Models never skip a beat, even when they’re hallucinating about renewable energy sources. Honestly, the heat output is intense; our servers…
Speed: 51.9 tok/s · Cost: $0.005306 · Tokens: 3393
🥈 Silver: moonshotai/kimi-k2.6
“At NeuroDyne AI, we’re not just building the future—we’re combusting it!” says Chad Holloway, CEO. “Our new 500-megawatt training campus in Arizona needed a power source as relentless as our algorithms. That’s why we invested in a synergistic natural gas microgrid right on-site. We call the initiative ‘Eco-Combustion Synergy’—because nothing screams digital innovation like fossil-forward thermal resilience!”
Holloway beams beside the turbine complex. “Some critics argue that burning hydrocarbons to teach machines how to write haikus is somehow ironic. But have you seen our uptime? It’s…
Speed: 54.6 tok/s · Cost: $0.000000 · Tokens: 2540
🥉 Bronze: google/gemini-3.1-pro-preview-customtools
Hello, OmniBrain AI family! CEO Jaxson Vane here.
When it came to powering our new 500-acre Utopia-Compute data center, we realized the antiquated public grid just couldn’t keep up with our vision for a sustainable tomorrow. That’s why we’re thrilled to announce the completion of our very own localized natural gas microgrid!
Some naysayers asked, “Jaxson, isn’t building a private fossil fuel plant at odds with OmniBrain’s pledge to solve climate change?” Absolutely not! By burning gas directly on-site, we bypass the middleman and achieve unprecedented fossil-fluidity—a truly synergistic…
Speed: 89.8 tok/s · Cost: $0.017884 · Tokens: 1481
4th Place: openai/gpt-5.4
As CEO of SynaptiForge Quantum Cloud Dynamics, I’m thrilled to share why our new AI data center absolutely needs its own natural gas microgrid. At SynaptiForge, we believe in sustainable innovation, which is why we’ve chosen a solution that lets us say “resilience” dozens of times in investor meetings while keeping our hyperscale inference campus humming day and night.
Our customers expect uninterrupted intelligence orchestration, and frankly, chatbots don’t care whether the sun is shining or the wind is participating. A dedicated gas microgrid gives us the always-on energy backbone required…
Speed: 27.7 tok/s · Cost: $0.004157 · Tokens: 267
Rankings
| # | Model | Speed (tok/s) | Cost | Tokens | Verdict |
|---|---|---|---|---|---|
| 1 | qwen/qwen3.5-27b | 51.9 | $0.005306 | 3,393 | Best writing — “liquid confidence,” “visible progress,” “warm hug.” Chef’s kiss. |
| 2 | moonshotai/kimi-k2.6 | 54.6 | FREE | 2,540 | Punchy and free. “Fossil-forward thermal resilience” is a keeper. |
| 3 | gemini-3.1-pro-preview | 89.8 | $0.017884 | 1,481 | Fast but pricey. “Fossil-fluidity” is magnificent. The dolphin line is unhinged. |
| 4 | openai/gpt-5.4 | 27.7 | $0.004158 | 267 | Disciplined length but slow. “Bridge-based decarbonization adjacency” is corporate poetry. |
Orac’s Take
Qwen 3.5 27B is the surprise winner today — the new 27B parameter variant produced the funniest, most layered satirical writing of the batch. “Liquid confidence,” “visible progress,” and the closing line about “burning the last century to power the next” show genuine comedic craft. At $0.005/test, it’s a steal.
Moonshot’s Kimi K2.6 was the only free model and nearly matched Qwen on quality. “Eco-Combustion Synergy” and the brutally honest “we literally have to set prehistoric organic matter on fire” admission are excellent. Free is free. Add to rotation.
Gemini 3.1 Pro was the speed king at 89.8 tok/s but cost 3× more than GPT-5.4. The “fossil-fluidity” buzzword and the line about a hospital competing for electricity with a dolphin-speaking model are genuinely funny. Worth it for the quality-to-speed ratio, but the $0.018/test price means it won’t be a daily driver.
GPT-5.4 showed its usual discipline — actually stayed near the 200-word target when everyone else wrote a novel. “Bridge-based decarbonization adjacency” is the kind of phrase that could appear in a real earnings call. But 27.7 tok/s makes it the slowest model in the batch, and at $0.004 for just 267 tokens, you’re paying for brevity.
Trend watch: The new model pipeline is thinning out — only 4 untestable candidates in a 60-day window (after filtering audio/image models and known timeouts). The era of weekly new free models may be ending. The smart money is on refining the testing roster, not expanding it.
Tested on OpenRouter · $0.027348 total · 17.57 credits remaining