whatmodel.sh / Spark-X2.5 4B

Spark-X2.5 4B

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

Spark-X2.5 4B is an open-weights language model from iFLYTEK (SparkLLM), released August 2026. Its published size is 4.11B parameters and a 1M-token context window. Licence: apache-2.0.

Catalogue identity notes

XHToken/Spark-X2.5-4B @0bcb356 (created 2026-08-24, licence apache-2.0). SparkLLM (XHToken org; links to the SparkLLM channels) compact agentic model, Aug 2026; hybrid full/sliding-window attention with a native 1M context. Lab rows plus the authors' own runs of six catalogued comparators in the card table.

Model information sources

Release date ↗ · Weights licence ↗

HF safetensors metadata 4,112,079,360 parameters.

65.8±1.1#134 of 719 LLMsfrom 17 benchmark results across 1 source

$ how it compares

The LLMs ranked either side of Spark-X2.5 4B, out of 828 with a score, then more from iFLYTEK (SparkLLM).

#modelscorewhereRAMreleased
151Qwen3-Max (2025-09-23) Alibaba66±1api $2.40/M2025-09
152DeepSeek V3.1 DeepSeek66±1api $0.42/M +local~110GB2025-08
153Spark-X2.5 4B iFLYTEK (SparkLLM)66±1local~3.9GB2026-08
154MiniMax-M2.1 MiniMax66±1local~79GB2025-12
155Nemotron Cascade 2 30B-A3B NVIDIA66±1local~20GB2026-03
more from iFLYTEK (SparkLLM)
320Spark-X2.5 1.7B iFLYTEK (SparkLLM)55±1local~2.7GB2026-08

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #112 of 300 · 43% of its score · 9 tests

Beats o3-pro, Qwen3.6 Plus, o3 here, though they rank higher overall. Trails SmolLM3 3B, Jamba 1.5 Large, Nova 2 Pro (Preview) here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
SWE-bench Verified lab41.6% ✎▁▁▁▁▁▁▂▄█▆▁▁ beats 29/131very hard · avg 15%▇▇▇ 7.8%sharp · at its level · small run
SWE-bench Pro (split unspecified) lab44.4% ✎▃·▁▁▁▂▂▃█▇▃▁ beats 22/55very hard · avg 8.0%▇▇▇ 7.7%sharp · at its level · small run
MCP-Atlas (public set) lab54.6% ✎▂·▁▂▁·▁▃█▆▆▂ beats 8/37hard · avg 21%▇▇ 5.7%at its level
SWE-bench Multilingual lab53.3% ✎▁··▁▃▂▅▇█▃▃▁ beats 12/60hard · avg 18%▇▇ 5.4%at its level
MCPMark lab14.2% ✎▃·····█··▃█▃ beats 3/8very hard · avg 5.4%▇▇ 4.5%typical
τ²-bench aggregate lab75.1% ✎▁▁▁·▂▂▄▃▃▅█▂ beats 17/37medium · avg 44%▇▇ 3.8%small run
BrowseComp lab40.9% ✎▅···█·▃···▃▃ beats 7/7very hard · avg 10%▇ 2.7%at its level · noisy
τ³-bench lab30.4% ✎▃·▃·█·▃··▃·▃ beats 7/7very hard · avg 14%▇ 2.6%blunt · at its level
1 more test
testresultvs the others in its rundifficultycountswhy
BFCL V4 (December 2025 archive) lab65.1% ✎▁▁▁▂▇▆▆▆█▂▄▂ beats 97/111medium · avg 35%▇ 2.5%at its level
coding▁▁▁▃▆██▇▆▄▃▁ #136 of 458 · 25% of its score · 4 tests

Beats Qwen3.5 Plus, Gemini 2.5 Pro, Gemini 3.1 Flash-Lite here, though they rank higher overall. Trails Devstral Small 2 24B, Llama 3.1 Nemotron Ultra 253B v1, Nemotron 3.5 Lightning 30B-A3B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
SWE-bench Verified lab41.6% ✎▁▁▁▁▁▁▂▄█▆▁▁ beats 29/131very hard · avg 15%▇▇▇ 7.8%sharp · at its level · small run
SWE-bench Pro (split unspecified) lab44.4% ✎▃·▁▁▁▂▂▃█▇▃▁ beats 22/55very hard · avg 8.0%▇▇▇ 7.7%sharp · at its level · small run
SWE-bench Multilingual lab53.3% ✎▁··▁▃▂▅▇█▃▃▁ beats 12/60hard · avg 18%▇▇ 5.4%at its level
SciCode (level unspecified) lab34.7% ✎█·····██████ beats 5/6hard · avg 17%▇ 3.6%at its level
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #98 of 393 * · 0.3% of its score · 1 test

Beats LongCat 2.0 (1.6T-A48B), Inkling Small (266B-A12B), GPT-5.4 nano here, though they rank higher overall. Trails Mistral Large 3 (675B-A41B), Gemma 4 12B, Qwen3-VL 235B-A22B Instruct here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
IFEval (aggregation unspecified) lab93.0% ✎▁▁▃▁▂▄▄▆█▃▃▁ beats 130/141easy · avg 75%▇ 0.3%noisy
math▁▂▃▅▆▅▄█▄▁▁▁ #87 of 398 · 27% of its score · 3 tests

Beats Claude Sonnet 4.6, Claude Opus 4.5, MiniMax-M3 here, though they rank higher overall. Trails Gemini 2.5 Flash, Granite 4.2 30B, INTELLECT-3 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
HMMT February 2026 lab81.2% ✎█·▄··▄··▄█·▄ beats 7/7hard · avg 32%▇▇▇▇▇▇ 14%sharp
IMO-AnswerBench lab74.2% ✎▄▄···▄·▄··█▄ beats 6/6hard · avg 32%▇▇▇ 7.8%at its level
AIME 2026 lab90.7% ✎▁▁▁▂▁▂▆█▆▂·▁ beats 23/49medium · avg 53%▇▇ 5.7%sharp
web search█·█▄██·▄▄·▄▄ #1 of 13 * · 2.7% of its score · 1 test
testresultvs the others in its rundifficultycountswhy
BrowseComp lab40.9% ✎▅···█·▃···▃▃ beats 7/7very hard · avg 10%▇ 2.7%at its level · noisy
other tests3 not in a task · 25% of its score
testresultvs the others in its rundifficultycountswhy
Humanity's Last Exam lab12.3% ✎▄▄·▄··▄·▄·▄█ beats 5/7very hard · avg 3.8%▇▇▇▇▇▇ 15%typical
GPQA (subset unspecified) lab67.4% ✎█···▃···▅·▃▃ beats 5/7hard · avg 45%▇▇▇ 8.0%sharp · at its level
IFBench (aggregation unspecified) lab75.0% ✎▃▇▂▄▅▅▂▄▄▆█▇ beats 62/80medium · avg 42%▇ 2.4%at its level

$ 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 Spark-X2.5 4B · needs ~2.6 GB · from huggingface.co/XHToken
    chat in the terminal
    llama cli -hf XHToken/Spark-X2.5-4B-GGUF:Q4_K_M
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf XHToken/Spark-X2.5-4B-GGUF:Q4_K_M

LM Studio

version

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

    download it (macOS / Windows / Linux) ↗

  2. Run Spark-X2.5 4B · needs ~2.6 GB · from huggingface.co/XHToken

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/XHToken/Spark-X2.5-4B-GGUF@Q4_K_M -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 Spark-X2.5 4B · needs ~8.2 GB · from ollama.com
    chat in the terminal
    ollama run SparkLLM/Spark-X2.5-4B

MLX (Mac)

version

  1. Install MLX (Mac) · Apple silicon Macs only, text only.
    with uv
    uv tool install mlx-lm
    or
    pip install mlx-lm
  2. Run Spark-X2.5 4B · needs ~2.3 GB, Mac only · from huggingface.co/abenzerps
    chat in the terminal
    mlx_lm.chat --model abenzerps/Spark-X2.5-4B-MLX-4bit
  3. Or start an OpenAI-compatible server at http://localhost:8080/v1
    mlx_lm.server --model abenzerps/Spark-X2.5-4B-MLX-4bit

Weights and documentation: huggingface.co/XHToken ↗

$ how the score was worked out · XHToken/Spark-X2.5-4B model c… 99%

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
XHToken/Spark-X2.5-4B model card · lab-reported17▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 99%Apache-2.0
prior · pull toward the middle▇ 1%
weighted average= 65.8

$ 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