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	Model	Elo	Samples	Task Success Rate	TSR 95% CI
1	1-2	Gemini 3.1 Flash Live Minimal	1096±2795% 区间 1068 至 1123	841	74.6%	69.4%/79.2%
2	1-4	Gemini 3.8 Live	1083±3095% 区间 1053 至 1113	482	93.2%	89.0%/95.8%
3	2-5	Gemini 3.1 Flash Live High	1063±2795% 区间 1036 至 1090	813	71.8%	66.4%/76.7%
4	3-6	GPT-Live-1 (Sol, low)	1053±2795% 区间 1026 至 1081	652	90.9%	87.2%/93.7%
5	3-6	GPT-Live-1 (Astra, medium)	1048±2895% 区间 1020 至 1076	610	87.4%	82.9%/90.8%
6	3-9	Cartesia Line (Default Cascaded System)	1027±3795% 区间 990 至 1064	190	77.5%	68.9%/84.3%
7	6-10	Grok Voice Think Fast 2.0 High	1011±2895% 区间 983 至 1038	552	94.6%	91.4%/96.7%
8	8	GPT-Realtime-1.5	1000锚点固定为 1000，作为本榜单 Elo 刻度的参照点。	811	85.1%	80.8%/88.6%
9	7-12	ElevenLabs Agents (Default Cascaded System)	993±2695% 区间 967 至 1020	773	90.5%	86.7%/93.4%
10	7-12	Gemini 3.8 Live Extended Thinking (High)	990±2795% 区间 963 至 1017	643	89.1%	84.7%/92.4%
11	8-12	GPT Realtime (Aug '25)	974±2695% 区间 948 至 1000	797	89.4%	85.2%/92.6%
12	9-12	Nova 2.0 Sonic (Mar 2026)	974±2695% 区间 948 至 1000	795	57.1%	51.1%/62.9%
13	13-19	GPT-Realtime-2.1 Minimal	937±2695% 区间 911 至 963	798	89.4%	85.2%/92.5%
14	13-20	GPT-Realtime-2 (High)	932±2695% 区间 906 至 958	764	89.8%	85.7%/92.9%
15	13-20	GPT-Realtime-2.1 High	928±2695% 区间 902 至 954	749	91.5%	87.7%/94.3%
16	13-20	GPT Realtime Mini (Oct '25)	917±2695% 区间 892 至 943	830	79.6%	74.6%/83.8%
17	13-20	GPT-Realtime-2 (Minimal)	916±2695% 区间 890 至 941	830	84.7%	80.3%/88.3%
18	13-20	Deepgram Voice Agent (Default Cascaded System)	914±2695% 区间 888 至 940	817	73.7%	68.5%/78.2%
19	13-20	Nova Sonic	913±2695% 区间 888 至 939	827	57.1%	51.6%/62.5%
20	14-20	Grok Voice Think Fast 1.0	908±2695% 区间 882 至 934	864	80.7%	75.8%/84.8%
21	21-24	Inworld Realtime (Default Cascaded System)	867±2695% 区间 841 至 893	811	69.9%	64.3%/75.0%
22	21-25	Qwen3.5 Omni Plus Realtime	859±2695% 区间 833 至 886	821	62.0%	56.4%/67.2%
23	21-25	Qwen Audio 3.0 Realtime Flash	850±4095% 区间 810 至 889	166	81.7%	70.1%/89.4%
24	21-25	Qwen3.5 Omni Flash Realtime	841±2695% 区间 814 至 867	847	29.1%	24.0%/34.8%
25	22-25	GPT-Realtime-2.1 Mini Minimal	840±2795% 区间 813 至 867	821	76.7%	71.1%/81.4%
26	26	Qwen Audio 3.0 Realtime Plus	775±4495% 区间 731 至 819	171	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	Gemini 2.5 Flash	Eleven Flash v2	Pre-registered
Cartesia Line	Cartesia Line	Ink	Gemini 2.5 Flash	Sonic	Pre-registered schemas; Arena execution
Inworld Realtime	Inworld Realtime API	Inworld STT 1	Gemini 2.5 Flash	Inworld TTS 1.5 Mini	Session-level