Audio8 ASR Infinite is a native streaming speech recognition model for Chinese and English from Edge0, released as a preview that delivers the transcription base. It offers a selectable audio clock of 80, 120 or 160 ms (one text token per clock step, i.e. 12.5, 8.3 or 6.25 decisions per second) and a configurable transcription delay of 240 to 560 ms; target_delay_ms must be an integer multiple of the chosen clock, and the card lists post-trained operating points such as 80 ms with 240/320/480/560 ms delay. The checkpoint has a native context of 30 seconds, but with an adapted vLLM build and a rolling KV cache (a 30 s window with exact RoPE re-basing) it can transcribe unlimited-length audio around the clock with bounded memory and latency. Architecturally it inherits the Voxtral realtime audio design and DSM-style streaming: a causal audio tower initialised from Voxtral Realtime 4B (32 layers, hidden 1280, 128 mel bins, sliding window 750), a new audio projector and frame-length embedding, and a Qwen2.5-3B-Instruct decoder and LM head (36 layers, hidden 2048), all trained; the main weights are about 4.09B bfloat16 parameters (8.17 GB). A separate semantic_vad_heads.safetensors file adds semantic VAD heads (8 classes, horizons of 0.5 to 3.0 s) intended to separate thinking pauses, stuttering and a real end of turn. In the card's 480 ms delay, 80 ms clock evaluation with greedy decode, it reports CER of 1.750 on aishell1/test and 2.893 on aishell4/test, against 16.795 and 16.456 for Voxtral-Mini-4B-Realtime-2602 and 12.927 and 14.677 for nemotron-3.5-asr-streaming-0.6b (measured at 560 ms); on LibriSpeech test.clean and test.other its WER is 3.042 and 6.808, behind Voxtral's 2.210 and 5.552 but ahead of nemotron's 3.353 and 7.140. Inference runs through a Docker compose stack with a web demo and a websocket realtime client, plus a Torch simulated-streaming decode example. It is released under Apache-2.0, an arXiv paper is listed as coming soon, and the formal release with frame-level semantic perception is still in progress.