lane0 runs GLM-5.3-Flash at 1,518 output tokens per second: about ten times the fastest provider on OpenRouter, on the same open model, with the same answers.
Below, six providers receive the same prompt and write the same program. Each lane moves at that provider’s published speed. lane0 finishes in 0.99 seconds. The fastest of the rest takes 10.3.
Why speed matters
Agents wait on every token. A coding agent writes thousands of tokens for each task, and someone is waiting on every one. At 145 tokens per second a 10,000-token task takes over a minute. At 1,518 it takes 6.6 seconds.
Speed changes what gets built. An agent that answers in seconds can try more, check its own work and retry before anyone notices. Long reasoning stops being a loading screen.
Same model, same answer
Every lane runs GLM-5.3-Flash. We don’t swap in a smaller model or trade quality for speed: the tokens you get are the ones GLM would produce anyway. Your prompts, evals and integrations carry over unchanged.
The numbers
| Provider | tok/s | 1,500 tokens |
|---|---|---|
| lane0 | 1,518 | 0.99 s |
| Fireworks | 145 | 10.3 s |
| Parasail | 141 | 10.6 s |
| Together | 132 | 11.4 s |
| Baseten | 86 | 17.4 s |
| Z.ai | 33 | 45.5 s |
Early access
We’re opening access to teams whose products wait on tokens. Write to hello@lanezero.ai with what you would run.