Launch HN: Tokenless (YC S26) – Automatic model switching to save money

Tokenless is pitching a very tempting promise: swap in its service, let it secretly test several artificial intelligence models behind the scenes, then keep the cheapest one that seems to be doing the job right. The sales pitch is pure catnip for companies drowning in AI bills: same results, half the price. But on Hacker News, the crowd did what it does best — grabbed the shiny claim, turned it over, and started poking at every seam.

The biggest mood in the comments was skeptical accountant energy. Several readers zeroed in on one uncomfortable question: if Tokenless is asking multiple AI systems to start working at once, how exactly is that cheaper than just picking one model yourself? One commenter basically translated the launch page into everyday language: this only really saves money in narrow situations, especially when the system can't rely on a recent memory of the conversation. Another went even blunter: if your own setup already remembers previous requests properly, why would paying to "fan out" work be cheaper at all?

Then came the classic startup-thread drama: is this actually new? One user immediately compared the idea to DigitalOcean's model synthesis, politely but unmistakably puncturing the "novel" vibe. Others questioned whether the product would just get stuck with one model for an entire session, which makes the magical switching sound a lot less magical. Even the more supportive voices had a side-eye ready: nice idea, they said, but the real issue is whether quality holds up — because being cheaper is only fun until the answers get worse. In other words, the launch promised a bargain, and the comments responded with the internet's favorite phrase: show the receipts.