whatmodel.sh / Qwen3.8-Flash-Next

Qwen3.8-Flash-Next

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Qwen3.8-Flash-Next is an open-weights language model from Alibaba, released August 2026. Its published size is 180B parameters (6B active per token) and a 262k-token context window. Licence: qwen-community-1.0.

75.4±1.0#36 of 719 LLMsfrom 14 benchmark results across 5 sources

$ how it compares

The LLMs ranked either side of Qwen3.8-Flash-Next, out of 828 with a score, then more from Alibaba.

#modelscorewhereRAMreleased
35DeepSeek V3.2 Speciale DeepSeek76±3comparison only
36DeepSeek V4 Flash Vision (Exp) DeepSeek76±2local~104GB2026-08
37Qwen3.8-Flash-Next Alibaba75±1local~59GB2026-08
38Qwen3.8-Omni-Flash Alibaba75±2api $0.23/M2026-09
39Gemini 3.1 Pro (Preview) Google75±1api $4.50/M2026-02
more from Alibaba
18Qwen3.8 Max (2.4T-A95B) Alibaba78±1api $3.00/M +local~380GB2026-08
34Qwen3.8 Max (0902) Alibaba76±1api $3.00/M2026-09
51Qwen3.7 Max Alibaba74±1api $3.75/M2026-05

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #40 of 300 · 32% of its score · 4 tests

Beats Gemini 3.7 Flash here, though they rank higher overall. Trails GLM-5.2, GLM-5.3-Flash, Grok 4.7 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arena Agent (October 2026 aggregate IPS)-0.0073 IPS ✎▁▁▁▄▂▃▇█▂▂▁▃ beats 17/49avg -0.312 IPS▇▇▇▇▇▇▇▇ 19%sharp
SWE-bench Multilingual lab81.0% ✎▁··▁▃▂▅▇█▃▃▁ beats 51/60hard · avg 29%▇▇ 5.7%typical
SWE-bench Pro public set, Qwen-refined lab62.5% ✎▂·····▂·▂█▃▅ beats 11/12hard · avg 18%▇▇ 4.9%at its level
DeepSWE v1.1 (resolved) lab58.7% ✎▂····▁▁▁▆██▆ beats 6/28hard · avg 15%▇ 1.5%at its level · noisy
coding▁▁▁▃▆██▇▆▄▃▁ #17 of 458 · 40% of its score · 6 tests

Beats GPT-5.6 Sol, GPT-5.5, MiMo-V2.6-Pro here, though they rank higher overall. Trails Grok 4.7 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arena WebDev (Code Arena)1637.5 ✎▁▂▄▃▅█▄▇▆▄▁▁ beats 107/119avg 981▇▇▇▇▇▇▇ 17%sharp
Luau Leetcode fully solved (Torpedo Software)76.4% ✎▁▁▁▁·▁▂▃▅▃█▇ beats 26/49hard · avg 18%▇▇▇ 6.6%at its level
SWE-bench Multilingual lab81.0% ✎▁··▁▃▂▅▇█▃▃▁ beats 51/60hard · avg 29%▇▇ 5.7%typical
SWE-bench Pro public set, Qwen-refined lab62.5% ✎▂·····▂·▂█▃▅ beats 11/12hard · avg 18%▇▇ 4.9%at its level
LiveCodeBench v6 (window unspecified) lab91.9% ✎▃▅█▅▅▃▅···▃█ beats 15/16medium · avg 35%▇▇ 4.0%typical
DeepSWE v1.1 (resolved) lab58.7% ✎▂····▁▁▁▆██▆ beats 6/28hard · avg 15%▇ 1.5%at its level · noisy
vision & documents▁▁▃▂▁▃▄▂▆▇█▄ #10 of 147 * · 10% of its score · 2 tests

Beats GPT-6.1 Sol, Claude Fable 5.1, Claude Sonnet 5.5 here, though they rank higher overall. Trails Claude Opus 4.7, Muse Spark 1.2 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
CharXiv (reasoning questions) lab84.6% ✎▄▂▄▆▄▄▆··▄█▂ beats 20/21medium · avg 39%▇▇▇▇ 10%typical
PureDocBench document parsing (Avg3 over Clean/Digital/Real)68.5% ✎▂·▁·▂▂▅█·▁▂▂ beats 10/22medium · avg 62%▇ 0.2%blunt · at its level
other tests5 not in a task · 30% of its score
testresultvs the others in its rundifficultycountswhy
GPQA Diamond lab91.7% ✎▁▁▁▁▂▄▅▅███▄ beats 284/310hard · avg 41%▇▇▇▇▇ 12%sharp
Humanity's Last Exam lab35.9% ✎▁▁▁▃▆▆▇██▆▁▃ beats 82/104very hard · avg 5.1%▇▇▇ 6.8%at its level
BRIDGE clinical text, zero-shot average (87 tasks)46.9% ✎▁▁▁▁▁▂▂▃▄▆█▃ beats 97/97hard · avg 32%▇▇ 5.4%blunt · at its level
IFBench (aggregation unspecified) lab81.3% ✎▃▇▂▄▅▅▂▄▄▆█▇ beats 77/80medium · avg 42%▇▇ 4.0%typical
RobloxQA v2.0 multiple choice (Torpedo Software)90.7% ✎▁▁▃▁█▄▃▃▆▂▂▂ beats 17/49easy · avg 78%▇ 2.2%maxed out

$ where to run it

llama.cpp · recommended

  1. Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.
    macOS / Linux
    curl -LsSf https://llama.app/install.sh | sh
    Windows
    irm https://llama.app/install.ps1 | iex

    other ways to install ↗

  2. Run Qwen3.8-Flash-Next · needs ~54.5 GB · from huggingface.co/unsloth
    chat in the terminal
    llama cli -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q2_K_XL
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q2_K_XL

LM Studio

  1. Install LM Studio · Open the app once before using the lms command.

    download it (macOS / Windows / Linux) ↗

  2. Run Qwen3.8-Flash-Next · needs ~54.5 GB · from huggingface.co/unsloth

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/unsloth/Qwen3.8-Flash-Next-GGUF@UD-Q2_K_XL -y
  4. Then run lms ls to see its name, and lms chat followed by that name

Ollama

  1. Install Ollama

    download it (macOS / Windows) ↗

    Linux
    curl -fsSL https://ollama.com/install.sh | sh
  2. Run Qwen3.8-Flash-Next · needs ~108.5 GB · from ollama.com
    chat in the terminal
    ollama run qwen3.8-flash-next:125b-a6b-q4_K_M

Official weights and documentation: huggingface.co/Qwen ↗

$ how the score was worked out · Qwen (Alibaba) vendor-reporte… 48%, LMArena leaderboard dataset 37%, Roblox LLM Leaderboard (Torpe… 9% and 2 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card) · lab-reported8▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 48%Factual vendor-reported score (not licensed content)
LMArena leaderboard dataset2▇▇▇▇▇▇▇▇▇▇▇ 37%CC BY 4.0
Roblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau Leetcode2▇▇▇ 9%MIT
BRIDGE Medical Leaderboard (Harvard Medical School / Brigham and Women's YLab)1▇▇ 5%MIT
How Far Is Document Parsing from Solved? PureDocBench: A Source-Traceable Benchmark across Clean, Degraded, and Real-World Settings (arXiv:2605.07492v2)1▇ 0%CC BY 4.0
prior · pull toward the middle▇ 0%
weighted average= 75.4
$ published speed measurements · 1

Reference results for the hardware, engine and settings below. They aren't verified speeds of the downloads above or predictions for your machine.

AMD Ryzen AI Max+ 395 (Strix Halo), Radeon 8060S iGPU; Beelink GTR9 Pro 128GBUD-IQ4_XS27.16 tok/sllama.cpp c841aee (build_number 1 as recorded), Vulkan (RADV) · llama-bench; tg128; 128 generated tokens; context depth 0; n_prompt=0 row; n_batch=2048; n_ubatch=512; flash_attn=1; type_k=f16; type_v=f16; n_gpu_layers=999; threads=16; one sequence; file label r10 repetitions

Author's own llama-bench CSV row (avg_ts mean over repetitions, stddev_ts retained). GPU string as recorded: Radeon 8060S Graphics (RADV STRIX_HALO). Unified memory; the guide reports ~124GiB OS-visible. Kernel/Mesa/power profile per the run folder notes. Not prompt processing or multi-request throughput. Curated in the guide's data/benchmarks.csv line 2 (2026-08-30, measured-local): Current-stack scout; 10 repeats; kernel 7.0.0-30, Mesa 26.1.7, desktop performance profile, DPM auto and recorded CPU-only background workload. Not a strict-clean speed headline or a same-condition regression comparison with older kernels. Artifact SHA-256 and exact effective flags are preserved in the raw bundle. Single ~93.7GB artifact capacity/direct-speed proof; a separate arithmetic smoke returned 56 after power restoration. No long-context, tools, vision, server or broad quality qualification. qwen-community-1.0 license; publisher counts 125B/6B active plus n-gram embeddings and MTP separately.

source ↗

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json