{
 "date": "2026-11-07",
 "model": "Kimi K3.1 Pro: 2.8T MoE, 92 MoE layers, 896 experts per layer, 16 routed per token; Q8_0 spine, MXFP4 experts (82,432 files, 1.45 TB)",
 "engine": "argonautlabsai/argodrive-ds4-kimi-k3.1, commit 6808130f7ff9fc94ff89e5535d308fc6b9589eb4",
 "drafter": "Qwen/Qwen3-0.6B-Base, revision da87bfb608c14b7cf20ba1ce41287e8de496c0cd, float16 on the same GPU; proposals of 2 to 15 tokens make blocks of up to 16 rows; gate 0.6 for raw text",
 "hardware": "one RTX PRO 6000 Blackwell (96 GB) in a 70 GiB VM; RAID0 of three NVMe with 79,365 experts; second NVMe with 41,078; Thunderbolt NVMe with 74,000",
 "settings": "documented defaults plus a third expert directory, DS4_K3_TRACE_HASH=1, -c 2048, greedy, one fresh process per run, monitoring off; raw mode except the chat rows, which use the chat template with DS4_K3_QWEN_CHAT=1 DS4_K3_QWEN_CONF=0.9 DS4_K3_QWEN_ADAPT=0.6,0.95,0.02,0.1",
 "prompts": {
  "record": "The three main financial statements are",
  "france": "The capital of France is",
  "public_512": "prompt-512-tokens.txt in the deltafin Kimi K3 benchmark",
  "chat": "a 286-token chat prompt and two general chat prompts, texts not published"
 },
 "output_check": "token ids and FNV-1a hashes of every position's logits compared with the plain run of the same prompt; 3416 logit vectors compared in the runs that fired blocks, 0 differ; plain token ids identical to the previous release's",
 "first_token": {
  "RN": {
   "first_token_s": 20.13,
   "prompt_tokens": 6,
   "prompt_pass_s": 3.03,
   "first_logit_hash": "2cec555e24621ed5"
  },
  "RO": {
   "first_token_s": 25.31,
   "prompt_tokens": 6,
   "prompt_pass_s": null,
   "first_logit_hash": "2cec555e24621ed5"
  },
  "ROb": {
   "first_token_s": 28.02,
   "prompt_tokens": 6,
   "prompt_pass_s": null,
   "first_logit_hash": "2cec555e24621ed5"
  },
  "RNb": {
   "first_token_s": 22.26,
   "prompt_tokens": 6,
   "prompt_pass_s": 3.11,
   "first_logit_hash": "2cec555e24621ed5"
  },
  "CN": {
   "first_token_s": 44.52,
   "prompt_tokens": 286,
   "prompt_pass_s": 24.66,
   "first_logit_hash": "804b714cd192bf31"
  },
  "CO": {
   "first_token_s": 49.07,
   "prompt_tokens": 286,
   "prompt_pass_s": null,
   "first_logit_hash": "804b714cd192bf31"
  },
  "COb": {
   "first_token_s": 49.33,
   "prompt_tokens": 286,
   "prompt_pass_s": null,
   "first_logit_hash": "804b714cd192bf31"
  },
  "CNb": {
   "first_token_s": 41.98,
   "prompt_tokens": 286,
   "prompt_pass_s": 24.67,
   "first_logit_hash": "804b714cd192bf31"
  },
  "LN": {
   "first_token_s": 51.5,
   "prompt_tokens": 1549,
   "prompt_pass_s": 34.17,
   "first_logit_hash": "6d2070a94dc341cb"
  },
  "LO": {
   "first_token_s": 57.38,
   "prompt_tokens": 1549,
   "prompt_pass_s": null,
   "first_logit_hash": "6d2070a94dc341cb"
  },
  "LOb": {
   "first_token_s": 56.91,
   "prompt_tokens": 1549,
   "prompt_pass_s": null,
   "first_logit_hash": "6d2070a94dc341cb"
  },
  "LNb": {
   "first_token_s": 51.31,
   "prompt_tokens": 1549,
   "prompt_pass_s": 34.07,
   "first_logit_hash": "6d2070a94dc341cb"
  }
 },
 "first_token_note": "fresh process, 2 generated, alternating; O = previous release (c0dcf69), N = this release; R = record prompt (6 tokens, raw), C = 286-token chat prompt, L = 1,549-token chat prompt; first token timed from process start the same way for both, by stamping every output line against a clock read just before exec; prompt-pass time for this release only, because the previous release read most of the spine inside its first prompt pass; the first token's logit hash is the same in every run of a prompt; these runs have the drafter off. With the drafter on, the sidecar loads during the spine read: in the drafter runs the spine came in at 2.6 to 3.2 GB/s instead of 3.6 to 3.7, the sidecar was ready after 19 to 24 s (2.5 to 2.7 s in the previous release), and the record prompt's first token came after 22.8 to 27.4 s instead of 20.0 to 20.3 s (the release's own counter, which agrees with the stamped clock to 0.04 s)",
 "failure_tests": "forced rejection of the first, a middle and the last draft, a mix, short proposals, end-of-text inside a block, a sidecar that dies, a malformed answer and a sidecar that cannot start: token ids and logits identical to plain in each",
 "not_established": [
  "comparison with Moonshot's own implementation",
  "exact match on other GPUs or drivers",
  "cancellation in the middle of a block and persistent-server recovery with the drafter on"
 ],
 "runs": [
  {
   "tag": "128P",
   "group": "record 128",
   "steady": 2.02,
   "steady_s": 62.78,
   "n": 127,
   "generated": 128,
   "wall": 86.071,
   "blocks": 0,
   "accepted": 0,
   "drafts": 0,
   "full": 0,
   "partial": 0,
   "zero": 0,
   "by_rows": "",
   "calls": 0,
   "call_ms": 0,
   "blk_tokens": 0,
   "ids": "reference: the other runs of this prompt are compared with this one",
   "hashes": "reference: the other runs of this prompt are compared with this one",
   "ids_vs_previous_release": "identical over 128"
  },
  {
   "tag": "128Q",
   "group": "record 128",
   "steady": 2.92,
   "steady_s": 43.56,
   "n": 127,
   "generated": 128,
   "wall": 73.176,
   "blocks": 19,
   "accepted": 97,
   "drafts": 107,
   "full": 14,
   "partial": 4,
   "zero": 1,
   "by_rows": "3:4 4:5 5:1 6:2 7:1 8:1 9:2 12:1 16:2",
   "calls": 30,
   "call_ms": 46.8,
   "blk_tokens": 116,
   "ids": "identical over 128",
   "hashes": "128 common, 0 differ"
  },
  {
   "tag": "128Qb",
   "group": "record 128",
   "steady": 2.9,
   "steady_s": 43.76,
   "n": 127,
   "generated": 128,
   "wall": 70.371,
   "blocks": 19,
   "accepted": 97,
   "drafts": 107,
   "full": 14,
   "partial": 4,
   "zero": 1,
   "by_rows": "3:4 4:5 5:1 6:2 7:1 8:1 9:2 12:1 16:2",
   "calls": 30,
   "call_ms": 47.4,
   "blk_tokens": 116,
   "ids": "identical over 128",
   "hashes": "128 common, 0 differ"
  },
  {
   "tag": "128Pb",
   "group": "record 128",
   "steady": 2.02,
   "steady_s": 62.89,
   "n": 127,
   "generated": 128,
   "wall": 86.269,
   "blocks": 0,
   "accepted": 0,
   "drafts": 0,
   "full": 0,
   "partial": 0,
   "zero": 0,
   "by_rows": "",
   "calls": 0,
   "call_ms": 0,
   "blk_tokens": 0,
   "ids": "identical over 128",
   "hashes": "128 common, 0 differ"
  },
  {
   "tag": "200P",
   "group": "record 200",
   "steady": 1.99,
   "steady_s": 100.25,
   "n": 199,
   "generated": 200,
   "wall": 123.597,
   "blocks": 0,
   "accepted": 0,
   "drafts": 0,
   "full": 0,
   "partial": 0,
   "zero": 0,
   "by_rows": "",
   "calls": 0,
   "call_ms": 0,
   "blk_tokens": 0,
   "ids": "reference: the other runs of this prompt are compared with this one",
   "hashes": "reference: the other runs of this prompt are compared with this one",
   "ids_vs_previous_release": "identical over 200"
  },
  {
   "tag": "200Q",
   "group": "record 200",
   "steady": 3.1,
   "steady_s": 64.14,
   "n": 199,
   "generated": 200,
   "wall": 93.396,
   "blocks": 24,
   "accepted": 164,
   "drafts": 174,
   "full": 19,
   "partial": 4,
   "zero": 1,
   "by_rows": "3:4 4:5 5:1 6:2 7:1 8:2 9:2 12:1 16:6",
   "calls": 35,
   "call_ms": 55.6,
   "blk_tokens": 188,
   "ids": "identical over 200",
   "hashes": "200 common, 0 differ"
  },
  {
   "tag": "200Qb",
   "group": "record 200",
   "steady": 3.11,
   "steady_s": 64.06,
   "n": 199,
   "generated": 200,
   "wall": 90.284,
   "blocks": 24,
   "accepted": 164,
   "drafts": 174,
   "full": 19,
   "partial": 4,
   "zero": 1,
   "by_rows": "3:4 4:5 5:1 6:2 7:1 8:2 9:2 12:1 16:6",
   "calls": 35,
   "call_ms": 54.7,
   "blk_tokens": 188,
   "ids": "identical over 200",
   "hashes": "200 common, 0 differ"
  },
  {
   "tag": "200Pb",
   "group": "record 200",
   "steady": 1.98,
   "steady_s": 100.33,
   "n": 199,
   "generated": 200,
   "wall": 123.798,
   "blocks": 0,
   "accepted": 0,
   "drafts": 0,
   "full": 0,
   "partial": 0,
   "zero": 0,
   "by_rows": "",
   "calls": 0,
   "call_ms": 0,
   "blk_tokens": 0,
   "ids": "identical over 200",
   "hashes": "200 common, 0 differ"
  },
  {
   "tag": "512P",
   "group": "record 512",
   "steady": 1.88,
   "steady_s": 272.2,
   "n": 511,
   "generated": 512,
   "wall": 295.63,
   "blocks": 0,
   "accepted": 0,
   "drafts": 0,
   "full": 0,
   "partial": 0,
   "zero": 0,
   "by_rows": "",
   "calls": 0,
   "call_ms": 0,
   "blk_tokens": 0,
   "ids": "reference: the other runs of this prompt are compared with this one",
   "hashes": "reference: the other runs of this prompt are compared with this one",
   "ids_vs_previous_release": "identical over 512"
  },
  {
   "tag": "512Q",
   "group": "record 512",
   "steady": 3.11,
   "steady_s": 164.1,
   "n": 511,
   "generated": 512,
   "wall": 191.289,
   "blocks": 46,
   "accepted": 443,
   "drafts": 459,
   "full": 40,
   "partial": 5,
   "zero": 1,
   "by_rows": "3:4 4:8 5:1 6:2 7:1 8:1 9:3 10:1 12:2 16:23",
   "calls": 68,
   "call_ms": 71.2,
   "blk_tokens": 489,
   "ids": "identical over 512",
   "hashes": "512 common, 0 differ"
  },
  {
   "tag": "512Qb",
   "group": "record 512",
   "steady": 3.11,
   "steady_s": 164.54,
   "n": 511,
   "generated": 512,
   "wall": 195.711,
   "blocks": 46,
   "accepted": 443,
   "drafts": 459,
   "full": 40,
   "partial": 5,
   "zero": 1,
   "by_rows": "3:4 4:8 5:1 6:2 7:1 8:1 9:3 10:1 12:2 16:23",
   "calls": 68,
   "call_ms": 70.6,
   "blk_tokens": 489,
   "ids": "identical over 512",
   "hashes": "512 common, 0 differ"
  },
  {
   "tag": "512Pb",
   "group": "record 512",
   "steady": 1.88,
   "steady_s": 271.8,
   "n": 511,
   "generated": 512,
   "wall": 295.324,
   "blocks": 0,
   "accepted": 0,
   "drafts": 0,
   "full": 0,
   "partial": 0,
   "zero": 0,
   "by_rows": "",
   "calls": 0,
   "call_ms": 0,
   "blk_tokens": 0,
   "ids": "identical over 512",
   "hashes": "512 common, 0 differ"
  },
  {
   "tag": "OP",
   "group": "public 512-token prompt, 200 generated",
   "steady": 1.53,
   "steady_s": 129.72,
   "n": 199,
   "generated": 200,
   "wall": 177.941,
   "blocks": 0,
   "accepted": 0,
   "drafts": 0,
   "full": 0,
   "partial": 0,
   "zero": 0,
   "by_rows": "",
   "calls": 0,
   "call_ms": 0,
   "blk_tokens": 0,
   "ids": "reference: the other runs of this prompt are compared with this one",
   "hashes": "reference: the other runs of this prompt are compared with this one",
   "ids_vs_previous_release": "identical over 200"
  },
  {
   "tag": "OQ",
   "group": "public 512-token prompt, 200 generated",
   "steady": 2.67,
   "steady_s": 74.64,
   "n": 199,
   "generated": 200,
   "wall": 126.215,
   "blocks": 13,
   "accepted": 185,
   "drafts": 185,
   "full": 13,
   "partial": 0,
   "zero": 0,
   "by_rows": "6:1 16:12",
   "calls": 14,
   "call_ms": 140.3,
   "blk_tokens": 198,
   "ids": "identical over 200",
   "hashes": "200 common, 0 differ"
  },
  {
   "tag": "OQb",
   "group": "public 512-token prompt, 200 generated",
   "steady": 2.64,
   "steady_s": 75.35,
   "n": 199,
   "generated": 200,
   "wall": 130.438,
   "blocks": 13,
   "accepted": 185,
   "drafts": 185,
   "full": 13,
   "partial": 0,
   "zero": 0,
   "by_rows": "6:1 16:12",
   "calls": 14,
   "call_ms": 150.0,
   "blk_tokens": 198,
   "ids": "identical over 200",
   "hashes": "200 common, 0 differ"
  },
  {
   "tag": "OPb",
   "group": "public 512-token prompt, 200 generated",
   "steady": 1.52,
   "steady_s": 130.92,
   "n": 199,
   "generated": 200,
   "wall": 178.842,
   "blocks": 0,
   "accepted": 0,
   "drafts": 0,
   "full": 0,
   "partial": 0,
   "zero": 0,
   "by_rows": "",
   "calls": 0,
   "call_ms": 0,
   "blk_tokens": 0,
   "ids": "identical over 200",
   "hashes": "200 common, 0 differ"
  },
  {
   "tag": "FRP",
   "group": "France, 17 generated",
   "steady": 1.95,
   "steady_s": 8.19,
   "n": 16,
   "generated": 17,
   "wall": 31.329,
   "blocks": 0,
   "accepted": 0,
   "drafts": 0,
   "full": 0,
   "partial": 0,
   "zero": 0,
   "by_rows": "",
   "calls": 0,
   "call_ms": 0,
   "blk_tokens": 0,
   "ids": "reference: the other runs of this prompt are compared with this one",
   "hashes": "reference: the other runs of this prompt are compared with this one"
  },
  {
   "tag": "FRQ",
   "group": "France, 17 generated",
   "steady": 2.17,
   "steady_s": 7.36,
   "n": 16,
   "generated": 17,
   "wall": 33.734,
   "blocks": 2,
   "accepted": 6,
   "drafts": 6,
   "full": 2,
   "partial": 0,
   "zero": 0,
   "by_rows": "3:1 5:1",
   "calls": 10,
   "call_ms": 40.3,
   "blk_tokens": 8,
   "ids": "identical over 17",
   "hashes": "17 common, 0 differ"
  },
  {
   "tag": "FRQb",
   "group": "France, 17 generated",
   "steady": 2.19,
   "steady_s": 7.32,
   "n": 16,
   "generated": 17,
   "wall": 37.739,
   "blocks": 2,
   "accepted": 6,
   "drafts": 6,
   "full": 2,
   "partial": 0,
   "zero": 0,
   "by_rows": "3:1 5:1",
   "calls": 10,
   "call_ms": 33.1,
   "blk_tokens": 8,
   "ids": "identical over 17",
   "hashes": "17 common, 0 differ"
  },
  {
   "tag": "FRQc",
   "group": "France, 17 generated",
   "steady": 2.2,
   "steady_s": 7.27,
   "n": 16,
   "generated": 17,
   "wall": 33.337,
   "blocks": 2,
   "accepted": 6,
   "drafts": 6,
   "full": 2,
   "partial": 0,
   "zero": 0,
   "by_rows": "3:1 5:1",
   "calls": 10,
   "call_ms": 31.4,
   "blk_tokens": 8,
   "ids": "identical over 17",
   "hashes": "17 common, 0 differ"
  },
  {
   "tag": "CP",
   "group": "286-token chat prompt, 200 generated",
   "steady": 1.94,
   "steady_s": 102.35,
   "n": 199,
   "generated": 200,
   "wall": null,
   "blocks": 0,
   "accepted": 0,
   "drafts": 0,
   "full": 0,
   "partial": 0,
   "zero": 0,
   "by_rows": "",
   "calls": 0,
   "call_ms": 0,
   "blk_tokens": 0,
   "ids": "reference: the other runs of this prompt are compared with this one",
   "hashes": "reference: the other runs of this prompt are compared with this one"
  },
  {
   "tag": "CQ",
   "group": "286-token chat prompt, 200 generated",
   "steady": 2.01,
   "steady_s": 98.91,
   "n": 199,
   "generated": 200,
   "wall": null,
   "blocks": 12,
   "accepted": 35,
   "drafts": 39,
   "full": 9,
   "partial": 3,
   "zero": 0,
   "by_rows": "3:6 4:2 5:2 6:1 9:1",
   "calls": 163,
   "call_ms": 12.9,
   "blk_tokens": 47,
   "ids": "identical over 200",
   "hashes": "200 common, 0 differ"
  },
  {
   "tag": "CQb",
   "group": "286-token chat prompt, 200 generated",
   "steady": 2.02,
   "steady_s": 98.67,
   "n": 199,
   "generated": 200,
   "wall": null,
   "blocks": 12,
   "accepted": 35,
   "drafts": 39,
   "full": 9,
   "partial": 3,
   "zero": 0,
   "by_rows": "3:6 4:2 5:2 6:1 9:1",
   "calls": 163,
   "call_ms": 12.9,
   "blk_tokens": 47,
   "ids": "identical over 200",
   "hashes": "200 common, 0 differ"
  },
  {
   "tag": "CPb",
   "group": "286-token chat prompt, 200 generated",
   "steady": 1.92,
   "steady_s": 103.87,
   "n": 199,
   "generated": 200,
   "wall": null,
   "blocks": 0,
   "accepted": 0,
   "drafts": 0,
   "full": 0,
   "partial": 0,
   "zero": 0,
   "by_rows": "",
   "calls": 0,
   "call_ms": 0,
   "blk_tokens": 0,
   "ids": "identical over 200",
   "hashes": "200 common, 0 differ"
  },
  {
   "tag": "GP1",
   "group": "general prompt 1",
   "steady": 1.98,
   "steady_s": 100.31,
   "n": 199,
   "generated": 200,
   "wall": null,
   "blocks": 0,
   "accepted": 0,
   "drafts": 0,
   "full": 0,
   "partial": 0,
   "zero": 0,
   "by_rows": "",
   "calls": 0,
   "call_ms": 0,
   "blk_tokens": 0,
   "ids": "reference: the other runs of this prompt are compared with this one",
   "hashes": "reference: the other runs of this prompt are compared with this one"
  },
  {
   "tag": "GQ1",
   "group": "general prompt 1",
   "steady": 2.13,
   "steady_s": 93.36,
   "n": 199,
   "generated": 200,
   "wall": null,
   "blocks": 34,
   "accepted": 79,
   "drafts": 98,
   "full": 27,
   "partial": 3,
   "zero": 4,
   "by_rows": "3:19 4:10 6:3 8:1 9:1",
   "calls": 120,
   "call_ms": 19.2,
   "blk_tokens": 113,
   "ids": "identical over 200",
   "hashes": "200 common, 0 differ"
  },
  {
   "tag": "GP5",
   "group": "general prompt 5",
   "steady": 1.86,
   "steady_s": 107.01,
   "n": 199,
   "generated": 200,
   "wall": null,
   "blocks": 0,
   "accepted": 0,
   "drafts": 0,
   "full": 0,
   "partial": 0,
   "zero": 0,
   "by_rows": "",
   "calls": 0,
   "call_ms": 0,
   "blk_tokens": 0,
   "ids": "reference: the other runs of this prompt are compared with this one",
   "hashes": "reference: the other runs of this prompt are compared with this one"
  },
  {
   "tag": "GQ5",
   "group": "general prompt 5",
   "steady": 3.03,
   "steady_s": 65.74,
   "n": 199,
   "generated": 200,
   "wall": null,
   "blocks": 19,
   "accepted": 171,
   "drafts": 174,
   "full": 18,
   "partial": 1,
   "zero": 0,
   "by_rows": "3:4 4:1 5:1 7:1 9:2 11:2 13:1 16:7",
   "calls": 28,
   "call_ms": 77.4,
   "blk_tokens": 190,
   "ids": "identical over 200",
   "hashes": "200 common, 0 differ"
  }
 ]
}