Act on this, mostly signal

SPEV: open-weight turn detection model with a free hosted API

What to do

Test SPEV for voice agent turn detection, since the hosted API is free through October 2026.

Why

Apache-2.0 30B omnimodal model on Qwen3-Omni, with reported much lower latency than OpenAI.

Read the source at huggingface.co Discussion

What the research found

SPEV is an Apache-2.0 open-weight 30B omnimodal turn-detection model on Qwen3-Omni with a LoRA adapter, with a free hosted API through October 2026 and reported much lower latency than OpenAI.

Claims and how well the evidence supports them

SPEV is a 30B omnimodal model built on Qwen3-Omni-30B-A3B with a LoRA adapter 98%
SPEV is Apache-2.0 licensed 99%
Hosted SPEV API on Sprag is free through October 2026 99%
SPEV p50 latency is 166 ms for 5-second audio versus 752 ms for OpenAI 98%

Evidence

huggingface.co https://huggingface.co/SpragAI/SPEV
news.ycombinator.com https://news.ycombinator.com/item?id=50089639

Actionability rated medium.