Speech Agent Arena LeaderboardArtificial Analysis
Compare Speech to Speech models by preference across blind, live voice conversations on topics like booking a dental appointment and ordering takeout. For more details, see overview and methodology, or visit Compare Models & Providers to compare Speech to Speech models across other benchmarks.
Range | Model | Elo | Samples | Task Success Rate | TSR 95% CI | |
|---|---|---|---|---|---|---|
| 1 | 1-2 |  Cartesia Managed Agents | 1111±2995% interval 1083 to 1140 | 685 | 91.7% | 88.3%/94.3% |
| 2 | 1-3 |  Gemini 3.1 Flash Live Minimal | 1089±2595% interval 1064 to 1114 | 990 | 74.6% | 69.4%/79.2% |
| 3 | 2-5 | Gemini 3.8 Live | 1074±2795% interval 1047 to 1101 | 606 | 93.2% | 89.0%/95.8% |
| 4 | 3-5 | Gemini 3.1 Flash Live High | 1062±2595% interval 1037 to 1086 | 957 | 71.8% | 66.4%/76.7% |
| 5 | 3-6 |  Grok Voice 3.0 | 1059±2895% interval 1031 to 1086 | 668 | 94.8% | 91.6%/96.8% |
| 6 | 5-8 |  GPT-Live-1 (Sol, low) | 1037±2595% interval 1012 to 1062 | 773 | 90.9% | 87.2%/93.7% |
| 7 | 6-8 | GPT-Live-1 (Astra, medium) | 1027±2595% interval 1002 to 1052 | 736 | 87.4% | 82.9%/90.8% |
| 8 | 6-11 |  Phonic 1 (High) | 1012±2695% interval 986 to 1038 | 683 | 92.3% | 88.9%/94.7% |
| 9 | 9 | GPT-Realtime-1.5 | 1000AnchorFixed at 1000 as the reference point for this board's Elo scale. | 921 | 85.1% | 80.8%/88.6% |
| 10 | 8-15 | Grok Voice 2.0 | 991±2595% interval 966 to 1016 | 669 | 94.6% | 91.4%/96.7% |
| 11 | 8-16 |  Eleven Realtime v1 Max | 990±2795% interval 964 to 1017 | 676 | 87.1% | 82.6%/90.6% |
| 12 | 9-16 | ElevenLabs Agents (Default Cascaded System) | 985±2495% interval 961 to 1010 | 893 | 90.5% | 86.7%/93.4% |
| 13 | 9-16 | Eleven Realtime v1 Mini | 981±2795% interval 954 to 1007 | 661 | 60.3% | 54.3%/66.0% |
| 14 | 10-16 | GPT Realtime (Aug '25) | 967±2495% interval 943 to 991 | 917 | 89.4% | 85.2%/92.6% |
| 15 | 12-16 | Gemini 3.8 Live Extended Thinking (High) | 966±2495% interval 941 to 990 | 773 | 89.1% | 84.7%/92.4% |
| 16 | 12-16 |  Nova 2.0 Sonic (Mar 2026) | 964±2495% interval 940 to 988 | 913 | 57.1% | 51.1%/62.9% |
| 17 | 17-21 | GPT-Realtime-2.1 Minimal | 933±2495% interval 909 to 957 | 932 | 89.4% | 85.2%/92.5% |
| 18 | 17-22 | GPT-Realtime-2.1 High | 931±2495% interval 907 to 955 | 887 | 91.5% | 87.7%/94.3% |
| 19 | 17-24 | GPT Realtime Mini (Oct '25) | 922±2495% interval 898 to 945 | 962 | 79.6% | 74.6%/83.8% |
| 20 | 17-24 | GPT-Realtime-2 (High) | 920±2495% interval 896 to 944 | 905 | 89.8% | 85.7%/92.9% |
| 21 | 17-24 | GPT-Realtime-2 (Minimal) | 918±2495% interval 894 to 941 | 980 | 84.7% | 80.3%/88.3% |
| 22 | 18-25 | Nova Sonic | 907±2695% interval 882 to 933 | 827 | 57.1% | 51.6%/62.5% |
| 23 | 19-25 | Grok Voice 1.0 | 903±2695% interval 877 to 929 | 864 | 80.7% | 75.8%/84.8% |
| 24 | 19-25 |  Deepgram Voice Agent (Default Cascaded System) | 901±2495% interval 877 to 925 | 955 | 73.7% | 68.5%/78.2% |
| 25 | 22-25 |  Inworld Realtime (Default Cascaded System) | 884±2495% interval 859 to 908 | 941 | 69.9% | 64.3%/75.0% |
| 26 | 25-29 |  Qwen Audio 3.0 Realtime Flash | 858±3495% interval 825 to 892 | 293 | 81.7% | 70.1%/89.4% |
| 27 | 26-29 | Qwen3.5 Omni Plus Realtime | 843±2595% interval 818 to 868 | 942 | 62.0% | 56.4%/67.2% |
| 28 | 26-29 | GPT-Realtime-2.1 Mini Minimal | 840±2595% interval 815 to 864 | 967 | 76.7% | 71.1%/81.4% |
| 29 | 27-29 | Qwen3.5 Omni Flash Realtime | 832±2595% interval 808 to 857 | 969 | 29.1% | 24.0%/34.8% |
| 30 | 29-30 | Qwen Audio 3.0 Realtime Plus | 798±3695% interval 762 to 834 | 316 | 77.8% | 66.9%/85.8% |