Hassan open sourced Tev1 0.8B, 50ms local on a Mac
The weights are out, the benchmarks are not, and Hassan says the small model is only for very simple classification.
Hassan (@nutlope) released the weights for Tev1 0.8B, the tiny Jev-like classifier he had shown running locally on his Mac, following the earlier post of the model itself. The download is public and he invited people to pull it.
Definitely not as good as Tev1 4B or Jev, but it runs entirely locally on a mac.
The one measured number attached to the model so far is latency. In his first post about it, Hassan showed the 0.8B running through Ollama on his Mac, classifying tasks at about 50ms end to end, and noted the video was not sped up. That is a single demo on his own machine, not a benchmark, and it says nothing about accuracy.
What he says it is not
Hassan is direct about the ceiling. He describes it as not as good as Tev1 4B or Jev, and says it can work decently well for very simple classification. He said when he first posted the model that weights and benchmarks were coming soon. The weights arrived. The benchmarks have not been posted, so there is currently no accuracy figure for the 0.8B against Jev, against the 4B, or against anything else.
Where it sits next to the 4B
The 0.8B follows tev1-4B-experimental, which Hassan fine-tuned on Qwen3.5 4B for $17 and released with the weights, the data recipe and a tutorial on training your own. That one is servable rather than local, listed on Together serverless at $0.042 per 1M input tokens and $0 per M output. The 0.8B changes the deployment story rather than the quality story. It runs on the laptop, with no API call and no per token bill.
For people building on it, treat the 50ms as a property of Hassan's Mac and his task set, not a spec. The useful test is whether your classification labels survive the drop from 4B to 0.8B, and until benchmarks land, the only way to find out is to run your own set against both. Hassan's own framing, very simple classification, is the sensible prior.
It's extremely fast: only ~50ms E2E latency. Video is not sped up!
Know what Jev is good at before you bet on it
One email a day. Under three minutes. What builders found out about Jev yesterday.
