Qwen3-235B-A22B-Thinking-2507 · RULER 4k(1M RULER · DCA · 260 samples/length · unit unspecified) 99.6 · 附件证据(ev-qg-completion-qwen-3-235b-a22b-thinking-2507-059)

Qwen3-235B-A22B-Thinking-2507 的 RULER 4k · 1M RULER · DCA · 260 samples/length · unit unspecified 官方出处。Qwen3-235B-A22B-Thinking-2507 · RULER · Full Attention · 99.6。见《Qwen3-235B-A22B-Thinking-2507 official Hugging Face model card》Processing Ultra-Long Texts / Long-Context Performance / Qwen3-235B-A22B-Thinking-2507 (Full Attention) / RULER 4k。证据 ev-qg-completion-qwen-3-235b-a22b-thinking-2507-059。

原文与定位

Qwen3-235B-A22B-Thinking-2507 · RULER 4k(1M RULER · DCA · 260 samples/length · unit unspecified),Qwen3-235B-A22B-Thinking-2507 · RULER · Full Attention配置,Reported score (unit unspecified) 99.6(具体成绩公布时间未确认)。见《Qwen3-235B-A22B-Thinking-2507 official Hugging Face model card》Processing Ultra-Long Texts / Long-Context Performance / Qwen3-235B-A22B-Thinking-2507 (Full Attention) / RULER 4k。 查看成绩条件

Long-Context Performance / 1M RULER | Model Name | Acc avg | 4k | 8k | 16k | 32k | 64k | 96k | 128k | 192k | 256k | 384k | 512k | 640k | 768k | 896k | 1000k | | Qwen3-235B-A22B-Thinking-2507 (Full Attention) | 95.4 | 99.6 | 100.0 | 99.5 | 99.6 | 99.1 | 100.0 | 98.8 | 98.1 | 96.1 | 95.2 | 90.0 | 91.7 | 89.7 | 87.9 | 85.9 | All models are evaluated with Dual Chunk Attention enabled. Since the evaluation is time-consuming, 260 samples are used for each length (13 sub-tasks, 20 samples for each). To avoid overly verbose reasoning, the thinking budget is set to 8,192 tokens.
官方材料
Qwen3-235B-A22B-Thinking-2507 official Hugging Face model card
定位
附件 · Processing Ultra-Long Texts / Long-Context Performance / Qwen3-235B-A22B-Thinking-2507 (Full Attention) / RULER 4k
材料版本
2026-11-17当前材料快照;完整读取Qwen3-235B-A22B-Thinking-2507模型卡正文、Performance表25个目标型号分数、1M RULER三配置表及图像;只录Thinking-2507两种长上下文注意力设置,排除旧Qwen3-235B-A22B行。表格和图像没有一般分数单位标记。
版本公布
未确认
披露时间
未确认 · 首发/后补未确认
采集时间
历史依据
暂无可靠历史版本依据
引用文本

Qwen3-235B-A22B-Thinking-2507 · RULER 4k(1M RULER · DCA · 260 samples/length · unit unspecified),Qwen3-235B-A22B-Thinking-2507 · RULER · Full Attention配置,Reported score (unit unspecified) 99.6(具体成绩公布时间未确认)。官方出处 Qwen3-235B-A22B-Thinking-2507 official Hugging Face model card(附件,2026-11-17当前材料快照;完整读取Qwen3-235B-A22B-Thinking-2507模型卡正文、Performance表25个目标型号分数、1M RULER三配置表及图像;只录Thinking-2507两种长上下文注意力设置,排除旧Qwen3-235B-A22B行。表格和图像没有一般分数单位标记。)Processing Ultra-Long Texts / Long-Context Performance / Qwen3-235B-A22B-Thinking-2507 (Full Attention) / RULER 4k,https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507。Benchlog https://benchlog.lekshome.cc/evidence/qwen-3-235b-a22b-thinking-2507/ev-qg-completion-qwen-3-235b-a22b-thinking-2507-059/

关联成绩与条件

Qwen3-235B-A22B-Thinking-2507 · RULER 4k 成绩与配置
指标与成绩配置与评测条件证据关系
Reported score (unit unspecified)99.6原文 99.6 · 单位 未标注
公布 未确认
Qwen3-235B-A22B-Thinking-2507 · RULER · Full Attention
设置
Official 1M RULER table Full Attention variant; Dual Chunk Attention enabled; 260 samples per length (13 subtasks × 20); thinking budget 8,192 tokens.
未知项
The Acc columns have no stated unit.
工具
工具条件未记录
推理预算
思考预算 8,192 tokens
框架
未记录
协议
未说明
条件
Qwen3-235B-A22B-Thinking-2507 (Full Attention); 1M RULER; Dual Chunk Attention enabled; 260 samples per length (13 subtasks × 20); thinking budget 8,192 tokens. 来源表头为 Acc,未注明单位,保留空单位。

直接支持该成绩

在排名中查看此成绩