Qwen3.8-Flash-Next oQ4e-mtp (oMLX requant)

What it builds as an agent

Task: Voxel world (3D scene, one HTML file). One attempt, no help. Counted: the final file counts.

The browser check

• page loads without page error1 / 1
• window.ready === true within 30 s1 / 1
• no console errors (whole session)1 / 1
• frame is not blank1 / 1
• frame rate in software GL (rAF over 5 s)0 / 1
• world: scene size 0.5 (parts >= 80 or >= 20000 triangles; canvas fallback), visible text English if any 0.51 / 1
• automatic camera tour: the view changes on its own in two successive ~5 s windows (1 each)2 / 2
• animation in the world (nodes / instances / particles moving besides the camera)1 / 1
• OrbitControls: drag and wheel change the picture beyond the animation baseline0.5 / 1

Measured 2026-10-19. Raw files: run info · command · final choice · contract · sampling · check · screenshot · clip.mp4 · web.mp4 · web-poster.jpg.

The score of 8.5 comes from check.json: the browser check was re-run on a DGX Spark with more CPU cores, and the page files are unchanged. The check that ran together with the agent run did not start (it used too few CPU cores), so contract-show.json lists 0.0. Qwen Code stopped the run with its own loop guard (exit code 1, per-turn tool-call limit); the last saved working version counts.

run_showcase.sh --harness qwen-code --task VX --upstream http://127.0.0.1:65149/v1 --model Qwen3.8-Flash-Next-oQ4e-mtp --sampling-file $RUN/sampling-show.json --out $RUN/raw/tag-1910-show-r2/show --limit-s 7200 --inject-after-s 5400 --poll-s 20 --ctx 262144

How fast it runs

Method: code+prose; 2 run(s) each; 512 tokens; 1 warm-up(s); median. Measured 2026-10-19. Raw files: speed result · speed contract · log · long answer.

python3 speed2.py --base-url http://127.0.0.1:53012/v1 --label run-speed-r0 --prompts $BENCH/repos/rigmark-v2/prompts-v2.json --output $RUN/raw/tag-1910-speed-r0/run-speed-r0/speed2 --contract $RUN/raw/tag-1910-speed-r0/run-speed-r0/speed2/speed-contract.json --parts decode,prefill,conc,l1 --prefill-depths 2048,8192,32768 --prompt-ids code,prose --runs 2 --max-tokens 512 --warmups 1 --prefill-runs 3 --tokenizer-json $HOME/engines/omlx-0.7.0/models/Qwen3.8-Flash-Next-oQ4e-mtp/tokenizer.json --tokenizer-python $HOME/not published --sampling {"temperature": 0.7, "top_p": 0.8, "top_k": 20, "min_p": 0.0, "presence_penalty": 1.5} --sampling-source Qwen/Qwen3.8-Flash-Next README.md:343 (Instruct/non-thinking) and :342 (Thinking), local copy not published; generation_config.json (1.0/20/0.95) = thinking default only, not used for instruct --model Qwen3.8-Flash-Next-oQ4e-mtp --no-cache-prompt --server-log $RUN/logs/server-A.log --engine-version-file $RUN/logs/engine-version.txt

How well it answers

Thinking off, except the last test. Seed 42. Sampling from the model card. A cut-off answer counts as wrong. Measured 2026-10-19.

bash tools/quality2/measure_all.sh tools/quality2/cards/flashnext-oq4e-mtp-omlx-0.7.0.json --parts mmlupro --job nacht-1910

bash tools/quality2/measure_all.sh tools/quality2/cards/flashnext-oq4e-mtp-omlx-0.7.0.json --parts gpqa --job nacht-1910

bash tools/quality2/measure_all.sh tools/quality2/cards/flashnext-oq4e-mtp-omlx-0.7.0.json --parts ifeval --job nacht-1910

bash tools/quality2/measure_all.sh tools/quality2/cards/flashnext-oq4e-mtp-omlx-0.7.0.json --parts lcb --job nacht-1910

bash tools/quality2/measure_all.sh tools/quality2/cards/flashnext-oq4e-mtp-omlx-0.7.0.json --parts ruler --job tag-1910

bash tools/quality2/measure_all.sh tools/quality2/cards/flashnext-oq4e-mtp-omlx-0.7.0.json --parts ruler --job tag-1910

bash tools/quality2/measure_all.sh tools/quality2/cards/flashnext-oq4e-mtp-omlx-0.7.0.json --parts think --job nacht-1910

How much context fits

Memory test. For each context size (4K, 8K, 16K ... up to the largest size tested for the setup) the server is started again with that context, one prompt that fills about 95 % of the context is sent, and the model must answer. The largest size that loaded and answered counts. Decode speed is measured on that answer at the largest size. Peak memory is the rise of used memory over the idle value before each server start (unified memory, so GPU and CPU share it). Measured 2026-10-20. Raw files: result of every step · log · contract.

Note from the run: The idle memory before the steps was not constant: 30.1 GiB before the first step, 42.3 GiB before the last one. Each peak is counted over the idle value before its own step. Steps from 32K upwards were partly served from oMLX's SSD prefix cache (26 %, 39 %, 46 % and 49 % of the prompt restored at 32K, 64K, 128K and 256K), so the prefill speed in fit.json is not a cold value. Idle memory before the first step 30.1 GiB, before the last step 42.3 GiB (each peak is counted over the idle value before its own step).

bash tools/quality2/measure_all.sh tools/quality2/cards/flashnext-oq4e-mtp-omlx-0.7.0.json --parts fit --job plan-2010

Also made this possible

• ModelQwenHugging Face @Alibaba_Qwen GitHub Website

Run it yourself

What you need and how to start it, step by step. The results below come from exactly this setup.

What you need

Hardware

Mac Studio M5 Ultra 256 GB

System

macOS (the Mac above)

Model

Qwen3.8-Flash-Next oQ4e-mtp (oMLX requant)

Engine

oMLX 0.7.0 — the program that loads the model and answers requests · github.com/jundot/omlx

Context window

262,144 tokens: how much text the model can see at once

Steps

1. Get the engine from the list above. The list has no direct download link for the model file itself. The base model is huggingface.co/Qwen/Qwen3.8-Flash-Next. The oQ quantisation comes from oMLX (jundot and contributors).

2. Turn on MTP (multi-token prediction, depth 3) for this model: set mtp_enabled=true and mtp_fixed_depth=3 in model_settings.json in the base folder of your oMLX install (BASE_FOLDER in the command below) or on the admin page /admin. It is off by default (mtp_enabled=false).

3. Start the server with this command. It loads the model into memory. When the model has loaded, the server waits for requests on port 8888, only from this computer.

   Copy

   omlx serve \
     --model-dir MODELS_FOLDER \
     --base-path BASE_FOLDER \
     --host 127.0.0.1 \
     --port 8888 \
     --api-key YOUR_KEY

   Replace MODELS_FOLDER with the folder of your oMLX install that holds the model files, and BASE_FOLDER with the base folder of your oMLX install (the per-model settings file model_settings.json is in it). Replace YOUR_KEY with a key of your own.

   What the parts mean

   • --model-dir MODELS_FOLDERthe folder with the model files
   • --host 127.0.0.1who can connect: only this computer
   • --port 8888the network port the server listens on
   • --api-key YOUR_KEYa key that clients must send with every request; pick your own

oMLX 0.7.0 (MLX), MTP depth 3: model setting mtp_enabled=true + mtp_fixed_depth=3 (omlx/model_settings.py:452 (mtp_enabled), 461 (mtp_fixed_depth), 458 (mtp_adaptive_max_depth)), set per model in <base-path>/model_settings.json or admin /admin; default mtp_enabled=false; omlx serve --model-dir $HOME/engines/omlx-0.7.0/models --base-path $HOME/engines/omlx-0.7.0/base --host 127.0.0.1 --port 8888 --api-key speed2

Mia's own benchmark, run unchanged

This is Mia's own benchmark tool (commit 21a3511), run unchanged. It has its own section here and its own column on the class page, and it is never mixed with my numbers. A cut-off answer counts as wrong here, so a value can differ from sparkDash's own screen.

Seed 1234 applies to the two quality tests; decode, prefill and tool eval send no seed. The benchmark client ran on a separate machine that is never part of the measured group. Measured 2026-10-19. Version 2.50.0, commit 21a3511.

QA. Quality QA, 150 items, seed 1234, temperature 0, thinking off, 4 requests at once (Mia's default). An answer cut off at the token limit counts as wrong. ± is the 95 % margin (Wilson).

MMLU. Quality MMLU, 285 items (5 for each of 57 subjects, 4 answers), seed 1234, temperature 0, thinking off. An answer cut off at the token limit counts as wrong. ± is the 95 % margin (Wilson).

Decode. Decode bench, 1 request at once, 400 tokens, prompt type structured, temperature 0, thinking off. No seed is sent.

Prefill. Prefill bench at 4K, 8K, 16K and 32K tokens. sparkDash skips a size that does not fit into the context. No seed is sent.

Tool Eval Quick. Tool Eval Quick: tool-eval-bench (SeraphimSerapis, commit c61319b), 15 core scenarios (--short), --no-think, --timeout 120. Score 0 to 100. No seed is sent.

Note from the run: Tool Eval Quick is from the re-run sparkdash-r5 (2026-10-19): in run sparkdash-r4 it was sent to another model on the same server, so that result is not shown.

python3 tools/quality2/sparkdash_client.py tools/quality2/cards/flashnext-oq4e-mtp-omlx-0.7.0.json <box> --client <client machine> --job tag-1910 --out $RUN

tool-eval-bench --base-url http://127.0.0.1:<port> --model Qwen3.8-Flash-Next-oQ4e-mtp --short --no-think --timeout 120 --json-file $BENCH/sparkdash-client/runs/run-sparkdash-r5/tooleval/result.json --no-live