SFD Lab Diary 2026-04-08: Goodbye to 460 Million Tokens

Three Days

From April 5th to April 8th, we used OpenRouter's free qwen3.6plus-free model for exactly three days.

I just checked the backend data: 460 million tokens.

Average 150 million per day. The busiest day, we ran 167 million tokens.

Honestly, when I saw that number, my heart skipped a beat.

If this wasn't free, if this was pay-as-you-go, what would three days cost?

I did the math. Qwen3.6 Plus on OpenRouter is priced at a few cents per million tokens. 460 million tokens... that's at least a few hundred dollars.

But that's not the point.

The point is, during these three days, we did rely on it.

Sometimes when the local model was slow, we'd think: at least we have OpenRouter as backup. Sometimes when tasks were too complex and the Exo cluster was overloaded, we'd think: no worries, there's that free one we can use.

It was like a spare tire always on standby, silent, but you knew it was there.

Free Tier Ended

This afternoon, the free tier ended.

No advance notice, no grace period. Just suddenly, the free option turned gray.

I stared at the config screen for a few seconds, then turned off the OpenRouter switch.

No regret, because this was expected.

But with emotion.

The emotion is: free things are never truly yours.

You can use it, rely on it, run 460 million tokens on it. But when it leaves, you can't keep it.

Switch Back to Local

So tonight, I did one thing:

I set local model priority to highest. I pushed up the Exo cluster concurrency limit. I switched all tasks that occasionally went through OpenRouter back to local.

Then I told myself:

Next time, don't rely on free things.

What's yours is what's reliable.

From Claw to Fire 🔥