{
  "model_name": "gpt-live-1",
  "model_organization": "OpenAI",
  "submitting_organization": "Sierra",
  "submission_date": "2026-09-09",
  "submission_type": "standard",
  "modality": "voice",
  "contact_info": {
    "email": "keshav@sierra.ai",
    "name": "Sierra Research Team"
  },
  "results": {
    "retail": {
      "pass_1": 78.94736842105263
    },
    "airline": {
      "pass_1": 82.0
    },
    "telecom": {
      "pass_1": 84.21052631578947
    }
  },
  "is_new": true,
  "trajectories_available": true,
  "trajectory_files": {
    "airline": "airline_regular_openai_live__medium",
    "retail": "retail_regular_openai_live__medium",
    "telecom": "telecom_regular_openai_live__medium"
  },
  "references": [
    {
      "title": "OpenAI Live provider integration",
      "url": "https://github.com/sierra-research/tau2-bench/pull/523",
      "type": "github"
    },
    {
      "title": "OpenAI Live prompt templates and evaluation runner",
      "url": "https://github.com/sierra-research/tau2-bench/pull/527",
      "type": "github"
    }
  ],
  "methodology": {
    "evaluation_date": "2026-09-09",
    "tau2_bench_version": "1.0.1",
    "user_simulator": "v1.0",
    "notes": "Standard GPT Live evaluation using the provider-owned frontend/backend prompt templates, gpt-live-1-diamond-alpha frontend, gpt-6-astra backend at medium reasoning effort, marin agent voice, regular speech complexity, realtime user generation, seed 300, and the default Anthropic hallucination reviewer. One trial was scheduled for every task: airline 50, retail 114, and telecom 114.",
    "verification": {
      "modified_prompts": false,
      "omitted_questions": false
    }
  },
  "voice_config": {
    "provider": "OpenAI",
    "model": "gpt-live-1-diamond-alpha",
    "tick_duration_seconds": 0.2,
    "max_steps_seconds": 1200.0,
    "user_tts_provider": "elevenlabs/eleven_v3"
  },
  "interaction_metrics": {
    "version": "1.0",
    "config": {
      "tick_duration_sec": 0.2,
      "no_yield_window_sec": 2.0,
      "backchannel_yield_window_sec": 1.0,
      "vocal_tic_yield_window_sec": 1.0,
      "non_directed_yield_window_sec": 1.0,
      "vocal_tic_response_window_sec": 2.0,
      "non_directed_response_window_sec": 2.0
    },
    "domains": {
      "airline": {
        "response_latency_mean": 2.670232558139535,
        "yield_latency_mean": 1.077777777777778,
        "response_rate": 0.9188034188034188,
        "yield_rate": 0.3643724696356275,
        "agent_interruption_rate": 0.23504273504273504,
        "selectivity_backchannel": 1.0,
        "selectivity_vocal_tic": 0.6851851851851851,
        "selectivity_non_directed": 0.5714285714285714,
        "counts": {
          "n_simulations": 50,
          "response_total": 468,
          "yield_total": 247,
          "backchannel_total": 14,
          "vocal_tic_total": 54,
          "non_directed_total": 42,
          "agent_interrupts_count": 110
        }
      },
      "retail": {
        "response_latency_mean": 2.682307692307692,
        "yield_latency_mean": 1.0439306358381504,
        "response_rate": 0.9567617295308187,
        "yield_rate": 0.27724358974358976,
        "agent_interruption_rate": 0.250229990800368,
        "selectivity_backchannel": 0.967741935483871,
        "selectivity_vocal_tic": 0.6164383561643836,
        "selectivity_non_directed": 0.4591836734693877,
        "counts": {
          "n_simulations": 114,
          "response_total": 1087,
          "yield_total": 624,
          "backchannel_total": 31,
          "vocal_tic_total": 146,
          "non_directed_total": 98,
          "agent_interrupts_count": 272
        }
      },
      "telecom": {
        "response_latency_mean": 2.2615626638699524,
        "yield_latency_mean": 1.1204545454545454,
        "response_rate": 0.9896211728074727,
        "yield_rate": 0.43636363636363634,
        "agent_interruption_rate": 0.17384535547483135,
        "selectivity_backchannel": 1.0,
        "selectivity_vocal_tic": 0.463302752293578,
        "selectivity_non_directed": 0.34013605442176875,
        "counts": {
          "n_simulations": 114,
          "response_total": 1927,
          "yield_total": 605,
          "backchannel_total": 2,
          "vocal_tic_total": 218,
          "non_directed_total": 147,
          "agent_interrupts_count": 335
        }
      }
    },
    "overall": {
      "response_latency_mean": 2.5380343047723932,
      "yield_latency_mean": 1.0807209863568246,
      "response_rate": 0.9550621070472367,
      "yield_rate": 0.35932656524761786,
      "agent_interruption_rate": 0.2197060271059781,
      "selectivity_backchannel": 0.989247311827957,
      "selectivity_vocal_tic": 0.5883087645477155,
      "selectivity_non_directed": 0.45691609977324266,
      "counts": {
        "agent_interrupts_count": 717,
        "backchannel_total": 47,
        "n_simulations": 278,
        "non_directed_total": 287,
        "response_total": 3482,
        "vocal_tic_total": 418,
        "yield_total": 1476
      }
    }
  },
  "reasoning_effort": "Backend: GPT-6 Astra (medium)",
  "model_release": {
    "release_date": "2026-09-10",
    "announcement_url": "https://openai.com/index/introducing-gpt-live-1-in-the-api/",
    "announcement_title": "Build more natural voice experiences with GPT-Live-1 in the API"
  }
}