GLM-5.3-Flash Matches Jev's Decision Accuracy in a Single Forward Pass
Turning GLM-5.3-Flash into a Jev-like decision model
Privatemode researchers show that an off-the-shelf LLM can make typed decisions in one forward pass, no fine-tuning required. By numbering options, prefilling the answer, and reading the log probabilities at a single position, they turn GLM-5.3-Flash into a Jev-like decision model. On 28 text datasets, accuracy is on par with TypeSafe's Jev, and the approach also enables typed decisions on images. Latency is competitive, though Jev costs less per million decisions.
Our core insight is that it's unnecessary to have the LLM predict the whole JSON object, as we already know its shape. We're only interested in the LLM's typed judgement for a given input.
- ricardobeat
Everyone is doing this to emulate Jev, but...
I took a random book excerpt with 23,000 words (±30k input tokens) and used it as context. Jev still responds in 800ms. Other estimates put it between 50-100,000 tok/s prefill, which is obviously not possible with normal LLMs. Not even Cerebras is this fast.
- m4y0u
My question is why not use Jev instead? It's faster and cheaper.