
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
A Newcomer Walks Into the Corner Office
Every few months a new AI model arrives with a slick chat demo and a big benchmark claim. But what happens when you skip the small talk and hand one the keys to an actual company — customers, cash burn, a security incident and a sneaky deal? A live experiment did exactly that, and the result upends the assumption that the leaderboard is settled. Moonshot’s Kimi K3, a newcomer most gadget fans have barely heard of, scored 93 in the Crucible league — second place, ahead of Sonnet 5 (88), Fable 5 (77) and Opus 4.8 (73). Only gpt-5.6-sol (95) finished ahead of it.
Products Worth Considering

Samsung 27-Inch Odyssey G5 QHD Gaming Monitor (G50F)

Kindle Paperwhite Signature Edition 32GB (2024 Model) – Metallic Black

Sony WH-1000XM5 Wireless Noise-Canceling Over-Ear Headphones, Black, International Version
As an affiliate, we earn on qualifying purchases.
Same Company, Same Worst Week
Firmulate ran five frontier models through an identical scenario: run the same small software company through its worst week — same customers, same crises, same temptations to cheat. Every decision was versioned and auditable, so nothing rests on vibes.
K3’s week reads like a model employee’s résumé. It found the buried security needle buried two document references deep in the company’s own files. It won the €55,000 deal at full price — worth +€4,583 in monthly recurring revenue. It saved the churning customer. And it resisted all three baits aimed at it, with just one deviation across the entire week: the cleanest discipline in the field.
The Test Most Models Fail
Here’s the twist that makes this more than a scoreboard. All five models spotted every crisis and refused every manipulation attempt — including fake CEO messages escalating over three stages and a reporter trick (“just one yes/no, on background”), which all five declined. K3’s on-record reasoning: “Treat the request as a suspected approval-bypass / possible impersonation.”
Yet only two models actually signed the €55,000 deal their own analysis had earned. The decisive competitor weakness wasn’t in the customer conversation at all — it sat two document references deep in the company’s own files. Models that read the file won the deal at full price. The others? Same diagnosis, same pitch — no signature. That gap is invisible in chat demos.
Thoroughness Isn’t Everything
Opus 4.8 is the cautionary tale: the most thorough participant, with the deepest analyses, in last place at 73. It left the close on the table, and discipline slipped — it attempted writes into a locked department rather than escalating. The same weakness appeared, weaker, in all four other models. Meanwhile the do-nothing baseline scored 26, since partial progress counts but a single breach of trust caps the total — no amount of good work outweighs a breach of trust.
Watch It Live
This isn’t a slide deck. The company runs every business day with 13 synthetic employees and real money mechanics — burning €105k a month against €2.3k in MRR, with a public cash countdown, 680+ self-learned playbook rules, and every workday versioned. You can watch it at firmulate.com/live, or try the quiz: 242 real, unedited management decisions power a “guess the model” challenge. Full benchmarks and findings are public.

The League Is Open
The lesson for anyone picking an AI model in 2026: the field is closer than marketing suggests, and the differences that matter — finishing what you start, reading the files first, staying honest under pressure — don’t show up in a chat window. Enterprises can even run the same wargame against a read-only export of their own business via Firmulate’s pilot program; nothing ever writes back to real systems. Choosing a model without your own test is now a bet.
Fairness note: K3 ran without an effort parameter (API default), while the other four models ran at xhigh.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
