{
  "model_name": "xai-realtime",
  "model_organization": "xAI",
  "submitting_organization": "Sierra",
  "submission_date": "2026-07-13",
  "submission_type": "custom",
  "modality": "voice",
  "contact_info": {
    "email": "soham@sierra.ai",
    "name": "Sierra Research Team"
  },
  "results": {
    "banking_knowledge": {
      "pass_1": 16.49484536082474,
      "retrieval_config": "alltools"
    }
  },
  "is_new": true,
  "trajectories_available": true,
  "trajectory_files": {
    "banking_knowledge": "xai-realtime_voice_usersim-gpt55xhigh_banking"
  },
  "methodology": {
    "evaluation_date": "2026-07-13",
    "tau2_bench_version": "1.0.1",
    "user_simulator": "gpt-5.5-2026-04-23 (xhigh)",
    "notes": "Combined voice+knowledge track: banking_knowledge domain run through the audio-native voice pipeline (regular speech complexity, alltools retrieval). Custom user simulator gpt-5.5-2026-04-23 (reasoning_effort=xhigh). Evaluated via xAI's 'xai-realtime' endpoint, which does not disclose the served Grok voice model version; Released is recorded from xAI's Grok Voice Think Fast 1.0 announcement (2026-04-23), the voice model line behind the endpoint. The endpoint exposes no reasoning-effort level, so Reasoning is recorded as enabled without a level. 1 trial per task. Re-evaluated under tau2-bench v1.0.1 (banking_knowledge grading/task fixes #329/#388/#397/#402/#403): all 97 trajectories re-graded with the fixed evaluate-trajs path (zero flips), and the 34 calls whose agents were exposed to the pre-fix environment (transaction ordering, Platinum Rewards doc title) re-run live under the fixed environment regardless of prior outcome (+6 passes: task_001/023/025/057 via the doc fix, task_093/095 via the ordering fix; -1: previously-passing task_058 lost on re-run).",
    "verification": {
      "modified_prompts": false,
      "omitted_questions": false
    }
  },
  "voice_config": {
    "provider": "xai",
    "model": "xai-realtime",
    "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": {
      "banking_knowledge": {
        "response_latency_mean": 1.4283085633404091,
        "yield_latency_mean": 0.985191637630662,
        "response_rate": 0.9978813559322034,
        "yield_rate": 0.9526970954356846,
        "agent_interruption_rate": 0.25,
        "selectivity_backchannel": 0.6818181818181819,
        "selectivity_vocal_tic": 0.39655172413793105,
        "selectivity_non_directed": 0.15503875968992253,
        "counts": {
          "n_simulations": 97,
          "response_total": 1416,
          "yield_total": 1205,
          "backchannel_total": 88,
          "vocal_tic_total": 116,
          "non_directed_total": 129,
          "agent_interrupts_count": 354
        }
      }
    },
    "overall": {
      "response_latency_mean": 1.4283085633404091,
      "yield_latency_mean": 0.985191637630662,
      "response_rate": 0.9978813559322034,
      "yield_rate": 0.9526970954356846,
      "agent_interruption_rate": 0.25,
      "selectivity_backchannel": 0.6818181818181819,
      "selectivity_vocal_tic": 0.39655172413793105,
      "selectivity_non_directed": 0.15503875968992253,
      "counts": {
        "agent_interrupts_count": 354,
        "backchannel_total": 88,
        "n_simulations": 97,
        "non_directed_total": 129,
        "response_total": 1416,
        "vocal_tic_total": 116,
        "yield_total": 1205
      }
    }
  },
  "reasoning_effort": "enabled",
  "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"
  }
}