Speech Agent Arena 排行榜 | Artificial Analysis

Speech Agent Arena 排行榜New Arena 概览与示例New

Speech Agent Arena 排行榜Artificial Analysis

通过实时语音对话的盲测偏好比较 Speech to Speech 模型，对话主题包括预约牙医和订购外卖。更多详情请参阅概览和方法论，或访问比较模型和提供商，在其他基准测试中比较 Speech to Speech 模型。

	Range	Creator	Model	Elo	Elo 95% CI	Samples	Task Success Rate	TSR 95% CI
1	1	Google	Gemini 3.1 Flash Live Preview - Minimal	1046	-30/30	768	74.6%	69.4%/79.2%
2	2-3	Google	Gemini 3.1 Flash Live Preview - High	1014	-29/29	763	71.8%	66.4%/76.7%
3	3	OpenAI	GPT-Realtime-1.5	1000	0/0	762	85.1%	80.8%/88.6%
4	4-6	OpenAI	GPT Realtime (Aug '25)	944	-27/27	747	89.4%	85.2%/92.6%
5	4-8	ElevenLabs	ElevenLabs Agents (Default Cascaded System)	937	-28/28	676	90.5%	86.7%/93.4%
6	4-11	Amazon Bedrock	Nova 2.0 Sonic (Mar 2026)	917	-27/27	718	57.1%	51.1%/62.9%
7	5-11	OpenAI	GPT-Realtime-2 (High)	914	-27/27	708	89.8%	85.7%/92.9%
8	5-11	OpenAI	GPT Realtime Mini (Oct '25)	912	-27/27	784	79.6%	74.6%/83.8%
9	6-13	SpaceXAI	Grok Voice Think Fast 2.0 High	908	-28/28	648	94.7%	91.5%/96.8%
10	6-13	OpenAI	GPT-Realtime-2.1 Minimal	896	-27/27	704	89.4%	85.2%/92.5%
11	6-13	OpenAI	GPT-Realtime-2.1 High	892	-27/27	656	91.5%	87.7%/94.3%
12	9-13	Deepgram	Deepgram Voice Agent (Default Cascaded System)	882	-27/27	730	73.7%	68.5%/78.2%
13	10-13	OpenAI	GPT-Realtime-2 (Minimal)	881	-27/27	789	84.7%	80.3%/88.3%
14	14-15	Amazon	Nova Sonic	852	-27/27	763	57.1%	51.6%/62.5%
15	14-15	SpaceXAI	Grok Voice Think Fast 1.0	839	-27/27	816	80.7%	75.8%/84.8%
16	16-19	Alibaba Cloud	Qwen3.5 Omni Plus Realtime	799	-27/27	779	62.0%	56.4%/67.2%
17	16-19	Alibaba Cloud	Qwen3.5 Omni Flash Realtime	797	-27/27	780	29.1%	24.0%/34.8%
18	16-19	OpenAI	GPT-Realtime-2.1 Mini Minimal	792	-28/28	742	76.7%	71.1%/81.4%
19	16-19	Inworld	Inworld Realtime (Default Cascaded System)	785	-29/29	683	69.9%	64.3%/75.0%
20	19-20	Alibaba Cloud	Qwen Audio 3.0 Realtime Flash	752	-39/39	168	81.7%	70.1%/89.4%
21	21	Alibaba Cloud	Qwen Audio 3.0 Realtime Plus	699	-40/40	186	77.8%	66.9%/85.8%

A Default Cascaded System is a provider’s standard voice-agent pipeline: its orchestration harness connects a speech-to-text model, an LLM, and a text-to-speech model. These entries use the provider’s default stack, rather than a custom pipeline optimized by Artificial Analysis.

Tool registration indicates whether tool definitions are pre-registered or supplied during session setup. As of July 2026:

Model	Harness	STT	LLM	TTS	Tool registration
Deepgram Voice Agent	Deepgram Voice Agent API	Nova-3	GPT-4o Mini	Aura-2	Session-level
ElevenLabs Agents	ElevenLabs Agents	Scribe v2 Realtime	GPT-4o Mini	Eleven v3	Pre-registered
Inworld Realtime	Inworld Realtime API	Inworld STT 1	Gemini 2.5 Flash	Inworld TTS 1.5 Mini	Session-level