{
  "claim": "Several numbers are circulating about Splash. Most of them come with a method; the loudest one does not. This is every Qwen3.8-27B throughput figure I could trace to a source, with the conditions attached to it and the memory bandwidth it implies.",
  "method": "Read from Inco's launch post (inco.ai/blog/splash/), the DFlash 2 post (inco.ai/blog/dflash2/), LM Studio's post (lmstudio.ai/blog/splash-engine) and the README of github.com/incoai/splash on 2026-09-19. The 'implied bandwidth' column is mine: claimed tokens/s divided by tokens per verification pass, times the bytes a pass reads — 16.73 GB for one lane at a short prompt, computed from the repository's own tensor shapes and its 4.5-bit pack format. Tokens per pass is Inco's published 4.80 unless a row says otherwise.",
  "source": "https://inco.ai/blog/splash/",
  "captured": "2026-09-19",
  "note": "The 144 tok/s figure is the one being quoted, and it is the only one with no written method attached: it is from the launch video, not the benchmark table. It is also the only one whose feasibility turns on which chip 'M5 Max' means — Apple sells two, at 460 and 614 GB/s, and only the larger one can produce it.",
  "columns": [
    { "key": "figure", "label": "figure", "align": "right" },
    { "key": "kind", "label": "" },
    { "key": "where", "label": "stated where" },
    { "key": "conditions", "label": "conditions" },
    { "key": "implied", "label": "implied bandwidth", "align": "right" }
  ],
  "rows": [
    {
      "figure": "74 tok/s",
      "kind": "Reported",
      "where": "launch post, Figure 2; repeated by LM Studio",
      "conditions": "48 GB M5 Pro, 16-core GPU · Qwen3.8-27B 4-bit · single request · short prompt · reasoning on at medium · 1,024-token output cap · P50 over selected SPEED-Bench coding prompts · measured over HTTP by Inco",
      "implied": "258 GB/s — 84% of the M5 Pro's 307"
    },
    {
      "figure": "54 tok/s",
      "kind": "Reported",
      "where": "launch post, Figure 2",
      "conditions": "same, at a 32K prompt. A pass now also reads 1.09 GB of int8 KV, so 17.82 GB in total. The drop from 74 is only 27%, which is small for a 32K context, and is the hybrid architecture showing: 48 of the 64 layers hold no KV at all",
      "implied": "200 GB/s — 65% of peak, if acceptance holds at 4.80"
    },
    {
      "figure": "170 tok/s",
      "kind": "Reported",
      "where": "launch post, Figure 5",
      "conditions": "aggregate across four concurrent short-prompt requests on the same M5 Pro. Four is not arbitrary: SPLASH_MAXIMUM_BATCH_WIDTH is 4, and the decode kernels exist in exactly the M8/M16/M24/M32 shapes that 1–4 lanes of 8 rows produce",
      "implied": "156 GB/s — half the bus. One weight read now serves four lanes, so throughput stops being a bandwidth question"
    },
    {
      "figure": "144 tok/s",
      "kind": "Reported, no method",
      "where": "launch video; quoted onward as the headline. Not in the benchmark table, not in the README, not in LM Studio's post",
      "conditions": "\"M5 Max\" — GPU bin unspecified. No prompt length, no concurrency, no reasoning setting, no output cap, no percentile",
      "implied": "502 GB/s — 109% of the 32-core M5 Max, 82% of the 40-core"
    },
    {
      "figure": "4.80 tokens",
      "kind": "Reported",
      "where": "DFlash 2 post, Table 4",
      "conditions": "mean acceptance length for this exact drafter on Qwen3.8-27B at block size 8, over GSM8K, MATH-500, HumanEval, MBPP and MT-Bench, at the model's default sampling. HumanEval is 4.39 and MBPP 4.79 — the coding-shaped ones, and the closest to what the Splash benchmark prompts with",
      "implied": "this is the input to every row above, not an output"
    },
    {
      "figure": "19 tok/s",
      "kind": "Reported",
      "where": "launch post, Figure 2",
      "conditions": "uzu on the same M5 Pro. The post states uzu's Qwen3.8-27B package has no draft, so it is one token per forward pass — the control condition for everything above",
      "implied": "280 GB/s, or 91% of peak, if uzu's 4-bit weights cost the same bytes as Splash's. My byte count caps a draft-free engine at 20.8 tok/s on this chip; uzu measures 19"
    },
    {
      "figure": "24 tok/s",
      "kind": "Reported, and awkward",
      "where": "launch post, Figure 2",
      "conditions": "Ollama on the same M5 Pro, same model",
      "implied": "354 GB/s at one token per pass — 115% of peak, which is impossible. So Ollama is not decoding one token per pass here: llama.cpp ships DFlash, and Qwen3.8-27B has a native MTP head that Inco's own Table 4 measures at 4.28"
    }
  ]
}