voice comparison metadata (9 entries)

gemini-3-1-flash-live-preview-thinking-high_google_2026-04-02/submission.json

@@ -0,0 +1,53 @@
+{
+  "model_name": "gemini-3.1-flash-live-preview-thinking-high",
+  "model_organization": "Google",
+  "submitting_organization": "Sierra",
+  "submission_date": "2026-04-02",
+  "submission_type": "standard",
+  "modality": "voice",
+  "contact_info": {
+    "email": "keshav@sierra.ai",
+    "name": "Sierra Research Team"
+  },
+  "results": {
+    "retail": {
+      "pass_1": 45.614035087719294
+    },
+    "airline": {
+      "pass_1": 64.0
+    },
+    "telecom": {
+      "pass_1": 21.929824561403507
+    }
+  },
+  "is_new": true,
+  "trajectories_available": true,
+  "trajectory_files": {
+    "airline": "airline_regular",
+    "retail": "retail_regular",
+    "telecom": "telecom_regular"
+  },
+  "reasoning_effort": "high",
+  "methodology": {
+    "evaluation_date": "2026-04-02",
+    "tau2_bench_version": "1.0.0",
+    "user_simulator": "v1.0",
+    "notes": "Evaluated using Gemini 3.1 Flash Live Preview with thinking = HIGH.",
+    "verification": {
+      "modified_prompts": false,
+      "omitted_questions": false
+    }
+  },
+  "voice_config": {
+    "provider": "Google",
+    "model": "gemini-3.1-flash-live-preview",
+    "tick_duration_seconds": 0.2,
+    "max_steps_seconds": 600.0,
+    "user_tts_provider": "elevenlabs/eleven_v3"
+  },
+  "model_release": {
+    "release_date": "2026-03-26",
+    "announcement_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-flash-live/",
+    "announcement_title": "Gemini 3.1 Flash Live: Google's latest AI audio model"
+  }
+}

gemini-3-1-flash-live-preview-thinking-minimal_google_2026-04-13/submission.json

@@ -0,0 +1,53 @@
+{
+  "model_name": "gemini-3.1-flash-live-preview-thinking-minimal",
+  "model_organization": "Google",
+  "submitting_organization": "Sierra",
+  "submission_date": "2026-04-13",
+  "submission_type": "standard",
+  "modality": "voice",
+  "contact_info": {
+    "email": "keshav@sierra.ai",
+    "name": "Sierra Research Team"
+  },
+  "results": {
+    "retail": {
+      "pass_1": 26.31578947368421
+    },
+    "airline": {
+      "pass_1": 42.0
+    },
+    "telecom": {
+      "pass_1": 17.543859649122805
+    }
+  },
+  "is_new": true,
+  "trajectories_available": true,
+  "trajectory_files": {
+    "airline": "airline_regular_gemini_gemini-3.1-flash-live-preview",
+    "retail": "retail_regular_gemini_gemini-3.1-flash-live-preview",
+    "telecom": "telecom_regular_gemini_gemini-3.1-flash-live-preview"
+  },
+  "reasoning_effort": "minimal",
+  "methodology": {
+    "evaluation_date": "2026-04-13",
+    "tau2_bench_version": "1.0.0",
+    "user_simulator": "v1.0",
+    "notes": "Evaluated using Gemini 3.1 Flash Live Preview with thinking = MINIMAL.",
+    "verification": {
+      "modified_prompts": false,
+      "omitted_questions": false
+    }
+  },
+  "voice_config": {
+    "provider": "Google",
+    "model": "gemini-3.1-flash-live-preview-thinking-minimal",
+    "tick_duration_seconds": 0.2,
+    "max_steps_seconds": 600.0,
+    "user_tts_provider": "elevenlabs/eleven_v3"
+  },
+  "model_release": {
+    "release_date": "2026-03-26",
+    "announcement_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-flash-live/",
+    "announcement_title": "Gemini 3.1 Flash Live: Google's latest AI audio model"
+  }
+}

gemini-live-2.5-flash_sierra_2026-03-03/submission.json

@@ -0,0 +1,52 @@
+{
+  "model_name": "gemini-live-2.5-flash-native-audio",
+  "model_organization": "Google",
+  "submitting_organization": "Sierra",
+  "submission_date": "2026-03-03",
+  "submission_type": "standard",
+  "modality": "voice",
+  "contact_info": {
+    "email": "victor@sierra.ai",
+    "name": "Sierra Research Team"
+  },
+  "is_new": true,
+  "trajectories_available": true,
+  "trajectory_files": {
+    "airline": "airline_regular_gemini_gemini-live-2.5-flash-native-audio",
+    "retail": "retail_regular_gemini_gemini-live-2.5-flash-native-audio",
+    "telecom": "telecom_regular_gemini_gemini-live-2.5-flash-native-audio"
+  },
+  "results": {
+    "airline": {
+      "pass_1": 30.0
+    },
+    "retail": {
+      "pass_1": 29.8
+    },
+    "telecom": {
+      "pass_1": 17.5
+    }
+  },
+  "voice_config": {
+    "provider": "Google",
+    "model": "gemini-live-2.5-flash-native-audio",
+    "tick_duration_seconds": 0.2,
+    "max_steps_seconds": 600,
+    "user_tts_provider": "elevenlabs/eleven_v3"
+  },
+  "methodology": {
+    "evaluation_date": "2026-03-03",
+    "tau2_bench_version": "1.0.0",
+    "user_simulator": "v1.0",
+    "notes": "Full-duplex audio-native evaluation using regular speech complexity (realistic audio conditions). 1 trial per task.",
+    "verification": {
+      "modified_prompts": false,
+      "omitted_questions": false
+    }
+  },
+  "model_release": {
+    "release_date": "2025-12-12",
+    "announcement_url": "https://cloud.google.com/blog/products/ai-machine-learning/gemini-live-api-available-on-vertex-ai",
+    "announcement_title": "Gemini Live API on Vertex AI"
+  }
+}

gpt-realtime-1-0_openai_2026-04-13/submission.json

@@ -0,0 +1,51 @@
+{
+  "model_name": "gpt-realtime-1.0",
+  "model_organization": "OpenAI",
+  "submitting_organization": "Sierra",
+  "submission_date": "2026-04-13",
+  "submission_type": "standard",
+  "modality": "voice",
+  "contact_info": {
+    "email": "victor@sierra.ai",
+    "name": "Sierra Research Team"
+  },
+  "results": {
+    "retail": {
+      "pass_1": 35.96491228070175
+    },
+    "airline": {
+      "pass_1": 36.0
+    },
+    "telecom": {
+      "pass_1": 19.298245614035086
+    }
+  },
+  "is_new": true,
+  "trajectories_available": true,
+  "trajectory_files": {
+    "airline": "airline_regular_openai_gpt-realtime-2025-08-28",
+    "retail": "retail_regular_openai_gpt-realtime-2025-08-28",
+    "telecom": "telecom_regular_openai_gpt-realtime-2025-08-28"
+  },
+  "methodology": {
+    "evaluation_date": "2026-03-27",
+    "tau2_bench_version": "1.0.0",
+    "user_simulator": "v1.0",
+    "verification": {
+      "modified_prompts": false,
+      "omitted_questions": false
+    }
+  },
+  "voice_config": {
+    "provider": "OpenAI",
+    "model": "gpt-realtime-1.0",
+    "tick_duration_seconds": 0.2,
+    "max_steps_seconds": 600.0,
+    "user_tts_provider": "elevenlabs/eleven_v3"
+  },
+  "model_release": {
+    "release_date": "2025-08-28",
+    "announcement_url": "https://openai.com/index/introducing-gpt-realtime/",
+    "announcement_title": "Introducing gpt-realtime"
+  }
+}

gpt-realtime-1.5_sierra_2026-03-03/submission.json

@@ -0,0 +1,52 @@
+{
+  "model_name": "gpt-realtime-1.5",
+  "model_organization": "OpenAI",
+  "submitting_organization": "Sierra",
+  "submission_date": "2026-03-03",
+  "submission_type": "standard",
+  "modality": "voice",
+  "contact_info": {
+    "email": "victor@sierra.ai",
+    "name": "Sierra Research Team"
+  },
+  "is_new": true,
+  "trajectories_available": true,
+  "trajectory_files": {
+    "airline": "airline_regular_openai_gpt-realtime-1.5",
+    "retail": "retail_regular_openai_gpt-realtime-1.5",
+    "telecom": "telecom_regular_openai_gpt-realtime-1.5"
+  },
+  "results": {
+    "airline": {
+      "pass_1": 40.0
+    },
+    "retail": {
+      "pass_1": 44.7
+    },
+    "telecom": {
+      "pass_1": 21.1
+    }
+  },
+  "voice_config": {
+    "provider": "OpenAI",
+    "model": "gpt-realtime-1.5",
+    "tick_duration_seconds": 0.2,
+    "max_steps_seconds": 600,
+    "user_tts_provider": "elevenlabs/eleven_v3"
+  },
+  "methodology": {
+    "evaluation_date": "2026-03-03",
+    "tau2_bench_version": "1.0.0",
+    "user_simulator": "v1.0",
+    "notes": "Full-duplex audio-native evaluation using regular speech complexity (realistic audio conditions). 1 trial per task.",
+    "verification": {
+      "modified_prompts": false,
+      "omitted_questions": false
+    }
+  },
+  "model_release": {
+    "release_date": "2026-02-23",
+    "announcement_url": "https://developers.openai.com/api/docs/models/gpt-realtime-1.5",
+    "announcement_title": "gpt-realtime-1.5 Model | OpenAI API"
+  }
+}

grok-voice-think-fast-1-0_xai_2026-04-21/submission.json

@@ -0,0 +1,52 @@
+{
+  "model_name": "grok-voice-think-fast-1.0",
+  "model_organization": "xAI",
+  "submitting_organization": "Sierra",
+  "submission_date": "2026-04-21",
+  "submission_type": "standard",
+  "modality": "voice",
+  "contact_info": {
+    "email": "victor@sierra.ai",
+    "name": "Sierra Research Team"
+  },
+  "results": {
+    "retail": {
+      "pass_1": 62.28070175438597
+    },
+    "airline": {
+      "pass_1": 66.0
+    },
+    "telecom": {
+      "pass_1": 73.68421052631578
+    }
+  },
+  "is_new": true,
+  "trajectories_available": true,
+  "trajectory_files": {
+    "airline": "xai-grok-voice-think-fast-1.0-airline",
+    "retail": "xai-grok-voice-think-fast-1.0-retail",
+    "telecom": "xai-grok-voice-think-fast-1.0-telecom"
+  },
+  "reasoning_effort": "enabled",
+  "methodology": {
+    "evaluation_date": "2026-04-21",
+    "tau2_bench_version": "1.0.0",
+    "user_simulator": "v1.0",
+    "verification": {
+      "modified_prompts": false,
+      "omitted_questions": false
+    }
+  },
+  "voice_config": {
+    "provider": "xAI",
+    "model": "grok-voice-think-fast-1.0",
+    "tick_duration_seconds": 0.2,
+    "max_steps_seconds": 1200.0,
+    "user_tts_provider": "elevenlabs/eleven_v3"
+  },
+  "model_release": {
+    "release_date": "2026-04-23",
+    "announcement_url": "https://x.ai/news/grok-voice-think-fast-1",
+    "announcement_title": "Grok Voice Think Fast 1.0"
+  }
+}

manifest.json

@@ -0,0 +1,15 @@
+{
+  "submissions": [],
+  "voice_submissions": [
+    "gpt-realtime-1.5_sierra_2026-03-03",
+    "gemini-live-2.5-flash_sierra_2026-03-03",
+    "xai-realtime_sierra_2026-03-03",
+    "grok-voice-think-fast-1-0_xai_2026-04-21",
+    "gemini-3-1-flash-live-preview-thinking-high_google_2026-04-02",
+    "gemini-3-1-flash-live-preview-thinking-minimal_google_2026-04-13",
+    "gpt-realtime-1-0_openai_2026-04-13",
+    "pine-realtime-1.0-preview_pine_2026-06-07",
+    "pine-realtime-1-0-preview_pine_2026-06-24"
+  ],
+  "legacy_submissions": []
+}

pine-realtime-1-0-preview_pine_2026-06-24/submission.json

@@ -0,0 +1,46 @@
+{
+  "model_name": "pine-realtime-1.0-preview",
+  "model_organization": "Pine",
+  "submitting_organization": "Pine",
+  "submission_date": "2026-06-24",
+  "submission_type": "custom",
+  "modality": "voice",
+  "contact_info": {
+    "email": "bojieli@gmail.com"
+  },
+  "results": {
+    "retail": {
+      "pass_1": 60.526315789473685
+    },
+    "airline": {
+      "pass_1": 56.00000000000001
+    },
+    "telecom": {
+      "pass_1": 26.31578947368421
+    }
+  },
+  "is_new": false,
+  "trajectories_available": true,
+  "trajectory_files": {
+    "airline": "airline_gpt41pure",
+    "retail": "retail_gpt41pure",
+    "telecom": "telecom_gpt41pure"
+  },
+  "methodology": {
+    "evaluation_date": "2026-06-24",
+    "tau2_bench_version": "1.0.0",
+    "user_simulator": "v1.0 (gpt-4.1 standard)",
+    "notes": "Standard v1.0 gpt-4.1 user simulator (content + turn-taking decision model) re-evaluation of pine-realtime-1.0-preview. Agent (agent3-default) and NL-assertions grader identical to the 2026-06-07 submission, which used a gpt-5.2 user simulator. Local agent3, pass^1, seed 300, regular speech.",
+    "verification": {
+      "modified_prompts": false,
+      "omitted_questions": false
+    }
+  },
+  "voice_config": {
+    "provider": "pinevoice",
+    "model": "agent3-default",
+    "tick_duration_seconds": 0.2,
+    "max_steps_seconds": 1200.0,
+    "user_tts_provider": "elevenlabs/eleven_v3"
+  }
+}

pine-realtime-1.0-preview_pine_2026-06-07/submission.json

@@ -0,0 +1,47 @@
+{
+  "model_name": "pine-realtime-1.0-preview",
+  "model_organization": "Pine",
+  "submitting_organization": "Pine",
+  "submission_date": "2026-06-07",
+  "submission_type": "standard",
+  "modality": "voice",
+  "contact_info": {
+    "email": "boj@19pine.ai",
+    "name": "Pine"
+  },
+  "results": {
+    "retail": {
+      "pass_1": 78.0701754385965
+    },
+    "airline": {
+      "pass_1": 70.0
+    },
+    "telecom": {
+      "pass_1": 84.21052631578947
+    }
+  },
+  "is_new": true,
+  "trajectories_available": true,
+  "trajectory_files": {
+    "airline": "airline",
+    "retail": "retail",
+    "telecom": "telecom"
+  },
+  "methodology": {
+    "evaluation_date": "2026-06-07",
+    "tau2_bench_version": "1.0.0",
+    "user_simulator": "v1.0",
+    "verification": {
+      "modified_prompts": false,
+      "omitted_questions": false
+    },
+    "notes": "pine-realtime-1.0-preview evaluated against all 278 tau2-bench voice tasks (50 airline, 114 retail, 114 telecom) at concurrency 8, one trial per task, regular speech complexity with realistic source and channel noise, openrouter/openai/gpt-5.2 as the simulated user. Per-domain pass_1 uses the full cohort as denominator; the simple mean across the three domains is 77.4%. Pass@1 with one retry: after the base run, every failed case was re-run exactly once and the re-run result is final (replace semantics), recovering transient user-simulator/inference failures so the numbers remain a true Pass@1."
+  },
+  "voice_config": {
+    "provider": "pinevoice",
+    "model": "pine-realtime-1.0-preview",
+    "tick_duration_seconds": 0.2,
+    "max_steps_seconds": 1200.0,
+    "user_tts_provider": "elevenlabs/eleven_v3"
+  }
+}

xai-realtime_sierra_2026-03-03/submission.json

@@ -0,0 +1,52 @@
+{
+  "model_name": "grok-voice-fast-1.0",
+  "model_organization": "xAI",
+  "submitting_organization": "Sierra",
+  "submission_date": "2026-03-03",
+  "submission_type": "standard",
+  "modality": "voice",
+  "contact_info": {
+    "email": "victor@sierra.ai",
+    "name": "Sierra Research Team"
+  },
+  "is_new": true,
+  "trajectories_available": true,
+  "trajectory_files": {
+    "airline": "airline_regular_xai_xai-realtime",
+    "retail": "retail_regular_xai_xai-realtime",
+    "telecom": "telecom_regular_xai_xai-realtime"
+  },
+  "results": {
+    "airline": {
+      "pass_1": 36.0
+    },
+    "retail": {
+      "pass_1": 38.6
+    },
+    "telecom": {
+      "pass_1": 40.4
+    }
+  },
+  "voice_config": {
+    "provider": "xAI",
+    "model": "grok-voice-fast-1.0",
+    "tick_duration_seconds": 0.2,
+    "max_steps_seconds": 600,
+    "user_tts_provider": "elevenlabs/eleven_v3"
+  },
+  "methodology": {
+    "evaluation_date": "2026-03-03",
+    "tau2_bench_version": "1.0.0",
+    "user_simulator": "v1.0",
+    "notes": "Full-duplex audio-native evaluation using regular speech complexity (realistic audio conditions). 1 trial per task.",
+    "verification": {
+      "modified_prompts": false,
+      "omitted_questions": false
+    }
+  },
+  "model_release": {
+    "release_date": "2025-12-17",
+    "announcement_url": "https://www.ainews.com/p/xai-launches-grok-voice-agent-api-for-real-time-ai-voice-apps",
+    "announcement_title": "xAI launches Grok Voice Agent API"
+  }
+}