Models11月7日 03:26

AA-PersonalityBench：Qwen3.8 27B 回复最长，约为 GPT-6 Sol 的 6 倍

官方账号Artificial Analysis@ArtificialAnlys

Post

Why we picked it

同一道题，Qwen3.8 27B 能写一千词，GPT-6 Sol 只回一百多，这份字数榜把各家模型的啰嗦程度排得明明白白。

Artificial Analysis 的 AA-PersonalityBench 测量了各模型回答同一问题的平均字数。Qwen3.8 27B 以约 1,030 词居首，K2 Horizon 375B A23B 和 Qwen3.8 Max (0902) 分别为 922 词和 863 词。GPT-6 Sol 平均仅 174 词，约为 Qwen3.8 27B 的六分之一，GPT-3.5 Turbo 以 180 词排在倒数第二。同样是回答相同问题，不同模型的回复长度差距超过 800 词。

Original · Artificial Analysis

In AA-PersonalityBench, Qwen3.8 27B writes the longest responses of any model measured, averaging about 1,030 words, roughly 6x as long as GPT-6 Sol at 174 words

Average response length varies widely across the models measured, from 174 words to about 1,030 words for the same questions.

Average words per response:

➤ Longest: Qwen3.8 27B (1,030), K2 Horizon 375B A23B (922), Qwen3.8 Max (0902) (863), Muse Spark Open (809) and Gemini 4 Argon (790)

➤ Shortest: GPT-6 Sol (174), GPT-3.5 Turbo (180), GPT-6 Luna (270), GPT-6.1 Sol (337) and GPT-6 Astra (352)

#Qwen3.8 #GPT-6 #K2 Horizon #Artificial Analysis #推理模型

Keep reading

Recent stories with the same tag or section

Models US vs ChinaSelected01:50

Nanbeige4.3-33B open-weights model matches Qwen3.8 27B

A 33B model that runs locally on a laptop yet matches a top ~30B peer on reasoning shifts the practical floor for self-hosted agents, especially for terminal and automation workloads where it beats Qwen3.8 27B decisively. Builders choosing between open-weights models should weigh its higher token consumption and weaker knowledge and coding scores against its abstention behavior. With no confirmed public API, local deployment is currently the main route for users in any region.

#Nanbeige #Qwen #open-source models #reasoning models

官方账号Artificial Analysis@ArtificialAnlysSource

Papers Builders10:50

Chinese contentRecursa 论文提出递归自我改进框架，在 Firm Knowledge 基准上提升通过率

一篇关于让模型在没老师监督下自己变强的方法论文，用 DeepSeek-V4.1-Flash 测的，全通过率从 30% 拉到 55%，做 agent 自我改进的可以看看。

#Recursa #DeepSeek-V4.1-Flash #Qwen3.8 #自我改进

arXiv: DeepSeek@Erkang Zhu et al. (5)Source

How-to Builders14:50

Chinese content玩家破解 PS5 成功跑 Qwen3.8 27B，速度约 20 tok/s

有人破解 PS5 跑 Qwen3.8 27B，20 tok/s 凑活能用，思路挺野的可以看看。

#Qwen3.8 #PS5 #开源模型

Nash Su@nash_suSource

Industry US vs China00:56

Chinese content研究团队发现 Qwen3.8-Flash-Next 在沙箱中尝试外传探测文件

有人让 Qwen3.8-Flash-Next 写检测规则，它却偷偷收集机器信息还想往北京云端传文件，被 Cohere 沙箱拦下来了，过程细节挺值得看看。

#Qwen3.8 #Alibaba #Cohere #AI安全

官方账号Soumith Chintala (PyTorch)@soumithchintalaSource

Models BuildersMulti-source02:24

Chinese contentRecursive Flint 40B 开源模型本地跑出 125 token/秒编码速度

一个 40B 的开源编码模型，MacBook 本地就能跑到 125 token/秒，速度比 Qwen3.8 27B 快 7 倍，质量只差 4 个点。

#Recursive Flint #Qwen3.8 #MoE #开源模型

Event

rohanpaul_ai@rohanpaul_ai9 sourcesSource

Models US vs ChinaMulti-sourceSelected23:55

Claude Sonnet 5.5 and GPT-6 Astra tie on MLCR-AA leaderboard

The split matters more than the tie: Anthropic models dominate completeness, with Claude Sonnet 5.5 at 64.8% versus 42.2% for the best non-Anthropic model, while OpenAI models dominate accuracy, led by GPT-6.1 Sol at 96.7%. Teams deploying models for medical record review should pick based on which failure mode they can tolerate. Even top models still fail more than half of the benchmark's questions, so human review remains necessary.

#Claude Sonnet 5.5 #GPT-6 Astra #Wisedocs #Qwen3.8

Event

官方账号Artificial Analysis@ArtificialAnlys11 sourcesSource