Best speech-to-speech models

Realtime voice models, scored on Big Bench Audio.

Qwen3.5 Omni Plus Realtime leads from Alibaba, ahead of 33 others.

ranked by big bench audio · 34 models · updated Aug 30

#	Model	Creator	Big Bench Audio
1	Qwen3.5 Omni Plus Realtime	Alibaba	0.99
2	Qwen Audio 3.0 Realtime Plus	Alibaba	0.99
3	Step-Audio R1.1 (Realtime)	StepFun	0.98
4	Gemini 3.1 Flash Audio Preview (Mar 2026) - High	Google	0.97
5	Grok Voice Think Fast 1.0	SpaceXAI	0.97
6	GPT-Realtime-2 High	OpenAI	0.97
7	Grok Voice Think Fast 2.0 High	SpaceXAI	0.97
8	Qwen Audio 3.0 Realtime Flash	Alibaba	0.96
9	GPT-Realtime-2.1 High	OpenAI	0.96
10	Grok Voice Agent	SpaceXAI	0.93
11	GPT-Realtime-2 Medium	OpenAI	0.93
12	Gemini 2.5 Flash Native Audio Dialog Thinking	Google	0.91
13	Nova 2.0 Sonic	Amazon	0.88
14	GPT-Realtime-2.1 Minimal	OpenAI	0.87
15	Deepslate Opal	Deepslate	0.85
16	GPT Realtime (Aug 2025)	OpenAI	0.83
17	GPT-Realtime-1.5	OpenAI	0.81
18	GPT-Realtime-2.1 Mini High	OpenAI	0.75
19	GPT-Realtime-2 Minimal	OpenAI	0.72
20	Gemini 3.1 Flash Audio Preview (Mar 2026) - Minimal	Google	0.71
21	Gemini 2.5 Flash Native Audio Dialog	Google	0.69
22	GPT-4o mini Realtime (Dec '24)	OpenAI	0.69
23	Higgs Realtime	Boson AI	0.69
24	GPT Realtime Mini (Oct 2025)	OpenAI	0.64
25	GPT-Realtime-2.1 Mini Minimal	OpenAI	0.63
26	Qwen3.5 Omni Flash Realtime	Alibaba	0.59
27	Qwen3 Omni 30B A3B Instruct	Alibaba	0.59
28	Qwen3 Omni 30B A3B Instruct (Realtime)	Alibaba	0.57
29	GPT 4o audio chatcompletions	OpenAI	0.54
30	Freeze-Omni	VITA-MLLM	0.33
31	Nemotron 3 VoiceChat (V1)	NVIDIA	0.27
32	PersonaPlex	NVIDIA	0.19
33	FLM-Audio	Cofe AI	0.16
34	Moshi	Kyutai	0.04

what big bench audio means

A spoken reasoning benchmark, scored from zero to one. Higher is better.