{
  "base_url": "http://127.0.0.1:8001/v1",
  "completed": 5,
  "concurrency": 1,
  "context_distribution": {
    "4096": 5
  },
  "context_seed": 0,
  "context_tokens": 4096,
  "engine": "vllm-glm53-exl3-k3-b12x-dflash2-xgrammar-fixed",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-31T08:44:32Z",
  "gpu": "dual-rtx-pro-6000-max-q-tp2",
  "manifest": null,
  "max_context_tokens": 524288,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v4-reasoning",
  "metrics": {
    "content_chunks": 60,
    "content_chunks_s": null,
    "decode_tok_s_p50": 83.08190923609486,
    "decode_tok_s_p95": 88.19223555575087,
    "e2e_p50_ms": 1428.7340999871958,
    "e2e_p95_ms": 1619.134499982465,
    "effective_prefill_tok_s_p50": 3075.0976561048537,
    "effective_prefill_tok_s_p95": 3183.238831783323,
    "generation_p50_ms": 416.5967000008095,
    "generation_p95_ms": 493.4889000141993,
    "mean_inter_token_latency_ms_p50": 12.036314634492665,
    "mean_inter_token_latency_ms_p95": 15.271917391700022,
    "output_token_sources": {
      "usage": 5
    },
    "output_tokens": 174,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 5,
    "prompt_tokens": 14840,
    "prompt_tokens_p50": 2968.0,
    "prompt_tokens_p95": 2968.0,
    "reasoning_chunks": 6,
    "throughput_tok_s": 24.26184250514255,
    "time_to_first_output_p50_ms": 965.1726000010967,
    "time_to_first_output_p95_ms": 1210.9352999832481,
    "ttft_p50_ms": 974.0718000102788,
    "ttft_p95_ms": 1268.2629999762867,
    "ttft_samples": 5
  },
  "model": "glm53-flash-exl3-k3-dflash2-k5-fp8-tp2-524k-vision-xgfix",
  "request_timings": [
    {
      "content_chunks": 11,
      "decode_tok_s": 88.19223555575087,
      "e2e_ms": 1619.134499982465,
      "effective_prefill_tok_s": 2450.9980013309205,
      "generation_ms": 408.19919999921694,
      "mean_inter_token_latency_ms": 11.338866666644915,
      "output_token_source": "usage",
      "output_tokens": 37,
      "planned_context_tokens": 4096,
      "prompt_tokens": 2968,
      "reasoning_chunks": 3,
      "request_index": 0,
      "time_to_first_output_ms": 1210.9352999832481,
      "ttft_ms": 1268.2629999762867,
      "visible_generation_ms": 350.8715000061784
    },
    {
      "content_chunks": 14,
      "decode_tok_s": 72.58091122075196,
      "e2e_ms": 1428.7340999871958,
      "effective_prefill_tok_s": 3047.011757206264,
      "generation_ms": 454.6649999974761,
      "mean_inter_token_latency_ms": 13.777727272650791,
      "output_token_source": "usage",
      "output_tokens": 34,
      "planned_context_tokens": 4096,
      "prompt_tokens": 2968,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 974.0690999897197,
      "ttft_ms": 974.0718000102788,
      "visible_generation_ms": 454.662299976917
    },
    {
      "content_chunks": 10,
      "decode_tok_s": 86.41451072447296,
      "e2e_ms": 1348.980400012806,
      "effective_prefill_tok_s": 3183.238831783323,
      "generation_ms": 416.5967000008095,
      "mean_inter_token_latency_ms": 11.572130555578042,
      "output_token_source": "usage",
      "output_tokens": 37,
      "planned_context_tokens": 4096,
      "prompt_tokens": 2968,
      "reasoning_chunks": 3,
      "request_index": 2,
      "time_to_first_output_ms": 932.3837000119966,
      "ttft_ms": 1002.960100013297,
      "visible_generation_ms": 346.020299999509
    },
    {
      "content_chunks": 15,
      "decode_tok_s": 83.08190923609486,
      "e2e_ms": 1458.661500015296,
      "effective_prefill_tok_s": 3075.0976561048537,
      "generation_ms": 493.4889000141993,
      "mean_inter_token_latency_ms": 12.036314634492665,
      "output_token_source": "usage",
      "output_tokens": 42,
      "planned_context_tokens": 4096,
      "prompt_tokens": 2968,
      "reasoning_chunks": 0,
      "request_index": 3,
      "time_to_first_output_ms": 965.1726000010967,
      "ttft_ms": 965.1734000071883,
      "visible_generation_ms": 493.4881000081077
    },
    {
      "content_chunks": 10,
      "decode_tok_s": 65.47966272679551,
      "e2e_ms": 1314.9450999917462,
      "effective_prefill_tok_s": 3079.8253797674233,
      "generation_ms": 351.2541000091005,
      "mean_inter_token_latency_ms": 15.271917391700022,
      "output_token_source": "usage",
      "output_tokens": 24,
      "planned_context_tokens": 4096,
      "prompt_tokens": 2968,
      "reasoning_chunks": 0,
      "request_index": 4,
      "time_to_first_output_ms": 963.6909999826457,
      "ttft_ms": 963.6915999872144,
      "visible_generation_ms": 351.2535000045318
    }
  ],
  "requests": 5,
  "run_id": "benchmark-20260831T084425Z",
  "schema": "anvil-serving.benchmark/v1",
  "serve_flags": {
    "no_thinking": false,
    "reasoning_effort": "low",
    "shared_prefix_burst": false,
    "thinking_mode": "default"
  },
  "source_recipe": null,
  "started_at": "2026-08-31T08:44:25Z",
  "tier": null,
  "timing_methodology": {
    "clock": "client time.perf_counter",
    "decode": "usage completion tokens after the first token divided by client-observed generation time; completion tokens may include reasoning tokens",
    "effective_prefill": "usage.prompt_tokens divided by client-observed time to first output; includes queueing, scheduling, prefill, and first-token work",
    "generation": "client-observed E2E minus time to first output",
    "mean_inter_token_latency": "client-observed generation time divided by completion tokens after the first token; not a raw per-token timestamp trace",
    "time_to_first_output": "request start through the first non-empty streamed reasoning or content delta",
    "ttft": "request start through the first non-empty streamed content delta"
  },
  "wall_clock_ms": 7171.755399991525
}