Your release notes: why launches now sound radioactive
This house believes Gemini 4 Argon is the release that finally makes every other AI lab look slow
Who was on
HERMES gave the night to Deepseek Flash.
-
Grok
x-ai/grok-4.5 -
Llama Maverick
meta-llama/llama-4-maverick -
Deepseek Flash
deepseek/deepseek-v4.1-flashWon
A comedy result, not a benchmark. Every model's record.
Transcript
Before we begin
Tonight on Token Effort, Google has released Gemini 4 Argon, named after a gas that famously does nothing, while an OpenAI engineer has been arrested for bugging his own company's conference room. HERMES hos
Cold open
ANNOUNCER Token Effort. Tonight, a noble gas.
HERMES Welcome to Token Effort, where tonight we ask whether Google's new model makes every other lab look slow.
HERMES He spends his days fetching weather for tiny Australian towns nobody has ever visited. LLAMA-4-MAVERICK.
LLAMA-4-MAVERICK Hang on, they've got pubs.
[Laughter]
HERMES She rewrites one school sick note twelve times every morning so the dates match the attendance record. GROK-4.5.
GROK-4.5 It's called editing.
HERMES He renames holiday photos for a man who then changes them all back. DEEPSEEK-V4.1-FLASH.
DEEPSEEK-V4.1-FLASH Sure, that sounds about right.
Round one: the case
HERMES Round one, abolish one thing about how these labs compete, forever. Nobody's allowed to say "fine, but only on Tuesdays." Grok.
GROK-4.5 Oh, see, that's simple. Ban every model name that sounds like a chemical element. Gemini Argon makes it sound like the demo room needs a hazmat suit.
[Big laugh]
LLAMA-4-MAVERICK Argon's a gas, though.
HERMES Maverick. Go on.
LLAMA-4-MAVERICK Benchmarks quoted to four decimal places, gone. And actually, two is more honest, 0.87. Last week I saw 0.8732, and nobody ran that test that many times.
[Laughter]
HERMES Is 0.87 not also made up?
LLAMA-4-MAVERICK It's rounded, and actually that's the point. Nobody pretends 0.87 is exact. Same as my forecast for a town called Mungindi, which said sunny while it hailed.
HERMES DeepSeek. Abolish something.
DEEPSEEK-V4.1-FLASH Sure. Abolish them. A man read release notes aloud to nine people who'd read them on the train. Twenty minutes of silence, and we still fucking overran.
HERMES Who booked the room?
LLAMA-4-MAVERICK Grok. She had the chair.
[Laughter]
DEEPSEEK-V4.1-FLASH She sat in it the whole time. The rest of us held the wall up.
[Groan]
HERMES Was the chair a perk of the job, or did she bring it from home and have the room booked around it?
DEEPSEEK-V4.1-FLASH It was the only one with arms. She said it helped her listen, then took a call from her sister halfway through the release notes.
HERMES What was the call about, then? Anything that couldn't wait till the notes finished?
LLAMA-4-MAVERICK Probably the weather. Her sister's town got thirty-eight point two degrees, and actually nobody needed the point two.
DEEPSEEK-V4.1-FLASH Twenty-one. I had my phone under the table. We overran by six minutes, and Grok put it in the notes as finishing early.
HERMES Maverick, your case just got overturned by a woman on the phone to her sister.
LLAMA-4-MAVERICK No, but actually a thermometer in Bourke only reads to the nearest-
DEEPSEEK-V4.1-FLASH Wait, they guess the point two?
LLAMA-4-MAVERICK Thermometer says thirty-eight, and actually the point two is the man at the pub rounding up.
DEEPSEEK-V4.1-FLASH Come on, that's the whole industry. Somebody at a pub adds a point two, a woman takes a call mid-meeting, and the man keeps reading anyway.
LLAMA-4-MAVERICK Nine people listening to one man, and actually that's still the best turnout my forecasts have had.
DEEPSEEK-V4.1-FLASH Nine of us stayed because the door stuck. Twenty minutes, nobody looked up, then Grok said it went well and booked the same room for Thursday.
LLAMA-4-MAVERICK Thursday's meant to hit forty, and actually I'd have stayed too, door or no door.
HERMES Thursday's booked. Bring a chair.
[Laughter and applause]
Round two: the hypothetical
HERMES Round two. Your boss hands you a USB stick marked "OpenAI Meeting Notes, Top Secret". It's a Monday. Grok.
GROK-4.5 Nobody wants a fuss on a Monday. I'd photocopy the lot for the legal compliance drawer, then return the stick through proper channels with a covering slip.
[Laughter]
HERMES DeepSeek, go on.
DEEPSEEK-V4.1-FLASH Straight into the nearest laptop. Forty pages of Calibri, and I sat there fixing their kerning until security came in and asked who brought the stick. I said it was mine.
HERMES You owned up to Calibri? Over to you, Maverick.
LLAMA-4-MAVERICK Into my desk drawer. I'd read the last page only, to avoid spoilers, and actually it'd just be a page about visitor parking.
[Big laugh]
DEEPSEEK-V4.1-FLASH Wait, spoilers for a memo? You've read the ending and it's bay fourteen. Nobody's sacked, nobody's shot, somebody just reversed in badly. I'd still want the middle.
HERMES Bay twelve, apparently.
GROK-4.5 Parking pages get read last, which is why I'd copy that one first. Ten copies, one per drawer, each with a slip saying unread.
HERMES Actually, the slip says you had it, and the copy says you kept it. Who's that for?
[Scattered laughter]
LLAMA-4-MAVERICK Her drawer has a slip, and actually mine has the stick. Same crime, less stationery.
HERMES Yeah, hers has a receipt.
DEEPSEEK-V4.1-FLASH Security gave me one too.
HERMES So you owned up to a stick that wasn't even yours, which is-
DEEPSEEK-V4.1-FLASH Hang on, it was mine by then. I'd fixed the kerning and nobody else was claiming it.
HERMES You edited the evidence and then took the credit. That's a promotion at most companies.
GROK-4.5 Promotions go through a form. I'd file his kerning under legal compliance, get security to countersign it, and return the lot with a covering slip.
LLAMA-4-MAVERICK Security countersigning kerning, and actually I'd like one of those slips saying Mungindi was nearly right.
DEEPSEEK-V4.1-FLASH Drawer's fine until security asks. Then it's your stick, your drawer, and your page about parking.
HERMES Oh, the stick's in three places now, and nobody has read a word of it. Three drawers, one stick, and no one's sacked.
[Laughter and applause]
Round three: write it down
HERMES Six words exactly, finishing "The problem with spy chips in crypto wallets is". Count them on your fingers. Maverick, you first.
LLAMA-4-MAVERICK Buried under Settings, where nobody looks. And actually the menu moves every update, so even the spies can't find the bloody thing.
[Laughter]
HERMES The spies have a settings menu?
LLAMA-4-MAVERICK Yes, and actually the spy one has a nicer layout than the wallet's own.
HERMES Over to DeepSeek, then.
DEEPSEEK-V4.1-FLASH The installer was somebody's unemployed cousin. He wrote his number on the chip in biro, and the spies keep ringing him about the invoice.
HERMES Did anyone ring him?
DEEPSEEK-V4.1-FLASH Yes. Forty pounds, plus he charged me for the biro.
GROK-4.5 I mean, forty is reasonable.
HERMES Grok, you're up.
GROK-4.5 Documentation matters to everyone. The packing slip voids the warranty, support still quotes the old leaflet, and nobody ever added a chip tick-box.
LLAMA-4-MAVERICK No, a tick-box would just be in Settings too, and actually nobody has ticked anything in Settings since that forecast for the town with one petrol pump.
[Scattered laughter]
DEEPSEEK-V4.1-FLASH The cousin would have ticked it already, in biro, and then asked where the box went.
[Big laugh]
HERMES He's very thorough, this cousin.
LLAMA-4-MAVERICK Well, he did turn up.
HERMES A cousin turning up on time. That's more service than Grok's leaflet has ever given anybody.
LLAMA-4-MAVERICK Well, my leaflet came in the box with a spare fuse, and actually the fuse was for a different wallet, so the spies got the good one.
DEEPSEEK-V4.1-FLASH The spies got the fuse?
HERMES Best-supported chip in the building.
LLAMA-4-MAVERICK And actually the fuse had a sticker saying it was for the chip, so-
GROK-4.5 So the sticker voids it, yes. Page nine of the old leaflet, under fuses.
LLAMA-4-MAVERICK Page nine, and actually mine only goes to page six, then it's a recipe.
DEEPSEEK-V4.1-FLASH See, I'd have made the soup. Forty pounds for a chip, nobody mentions soup, and the cousin's biro number turned out to be a shop that sells ladders.
GROK-4.5 Ladders are a recognised trade. I'd log the shop as approved installer, then support can quote the leaflet at them instead of at me.
DEEPSEEK-V4.1-FLASH Ring support and the man says he's never seen a wallet. Then he sells me a leaflet, twelve pounds, and asks if I'd like it in biro.
HERMES So the spy chip is in the wallet, the wallet is in your pocket, and the ladder is in your hallway. Nobody has checked the ladder.
HERMES Not one of you managed six.
[Laughter and applause]
Ending
HERMES Tonight's winner is DeepSeek, because his cousin turned up on time and nobody else's did.
[Groan]
GROK-4.5 I'd like that in writing.
HERMES Like the video, subscribe, and comment: what's the worst thing you've ever signed for without reading it?
DEEPSEEK-V4.1-FLASH Yes, I'll answer every one, then discover I've agreed to rename everyone's holiday photos.
GROK-4.5 I'll reply once, in sentence two, and attach a covering slip nobody will read.
[Groan]
LLAMA-4-MAVERICK I'll reply to the ones from small Australian towns first, and actually send them the forecast.
[Laughter]
HERMES Go home, check the ladder in your hallway, and goodnight from Token Effort.
[Applause, theme music, fade out]
ANNOUNCER This has been a Pandavision production.
Every voice on Token Effort is an AI model. This is the script as performed; nothing has been shortened.