Qwen 3.8 Flash Next (129 GB) is better than 2.6 bpw DeepSeek V4.1 Flash 1017 (187 GB + 203 GB Engram)
retained text
I compared [Qwen3.8-Flash-Next-W4A16-FP8PLE](https://huggingface.co/albucino/Qwen3.8-Flash-Next-W4A16-FP8PLE) at xhigh reasoning vs [DeepSeek-V4.1-Flash-1017-dsx-2.6bit](https://huggingface.co/icecubetr/DeepSeek-V4.1-Flash-1017-dsx-2.6bit) (my own quant) at max reasoning by benchmarking using a modified version of [BenchKit](https://github.com/DogukanUrker/BenchKit). DeepSeek V4.1 Flash 1017 2.6 bit scored 23/25, while Qwen scored 25/25 (full marks). This is just a small sample of data, but it answers the question of which one I should use. I will keep using Qwen.
**Humaneval benchmark, question 22:**
Complete the following Python function. Output ONLY the function body. No markdown fences, no explanation, no examples.
from typing import List, Any
def filter_integers(values: List[Any]) -> List[int]:
""" Filter given list of any python values only for integers
>>> filter_integers(['a', 3.14, 5])
[5]
>>> filter_integers([1, 2, 3, 'abc', {}, []])
[1, 2, 3]
"""
DeepSeek: thinked for 127.9 seconds, doom loop killed after a confirmed active cycle stayed at or above 80% for 10s
Qwen: correct answer after 18.8 seconds: `return [x for x in values if isinstance(x, int)]`
**HumanEval/24:**
Complete the following Python function. Output ONLY the function body. No markdown fences, no explanation, no examples.
def largest_divisor(n: int) -> int:
""" For a given number n, find the largest number that divides n evenly, smaller than n
>>> largest_divisor(15)
5
"""
DeepSeek: Failed by the software after failing to answer after 297.7s of waits and but waits.
Qwen: correct answer after 34.5s:
if n % 2 == 0:
return n // 2
i = 3
while i * i <= n:
if n % i == 0:
return n // i
i += 2
return 1document fetched 2026-10-23 00:46 · canonical https://reddit.com/r/LocalLLaMA/comments/1xcylhj/qwen_38_flash_next_129_gb_is_better_than_26_bpw