LlamaFactory
Qwen3-235B-A22B-Instruct-2507 · 快照 887b850813d8- 模板
- qwen3_nothink
- 训练阶段
- SFT / DPO / KTO / ORPO / SimPO
- 微调方式
- 全参数 / 冻结/部分参数 / LoRA / QLoRA
- 模态
- text
README 的 Qwen3 模板说明思考与非思考模式需选择对应 template;本档案使用 qwen3_nothink,不推断其他运行参数。
选择要核对的信息,直接进入对应章节;各项事实均保留版本、条件和来源。
5/14 项有公开值 · 当前版本 Qwen3-235B-A22B-Instruct-2507 · 数据截止 2026-09-17
页面只展示已审核 Claim;待审核数据不会被当作公开事实。只展示已审核事实;未公开、冲突和过期状态不会被补猜。
只判断选型资料是否足够,不评价模型能力高低。
已审核的公开资料尚不能确认 API 或本地部署路径,需要补充接入证据。
Qwen3-235B-A22B-Instruct-2507 是 阿里千问 / Qwen 的 Qwen3 模型,年鉴当前记录 1 个版本,最新公开记录为 Qwen3-235B-A22B-Instruct-2507。
权重为开放权重,代码为未披露,训练数据为未披露,部署方式为可自行部署;这些维度分别判断,不能合并成单一“开源/闭源”结论。
当前公开快照的上下文窗口为 暂无已审核记录,架构为 未公开;这里只使用通过审核且来源允许公开采集的 Claim。
7 个维度暂无可公开投影的审核记录
补充来源或线索会进入投稿队列,经编辑核验后才会成为公开事实。当前来源记录 2 个。
技术事实、业务价值、企业治理和证据质量分区呈现
厂商、家族、API ID、日期与生命周期
预训练、SFT、偏好优化与安全对齐
教师/学生、基础模型、轨迹和限制
权重、代码、数据、许可证、部署与 API
测试集、条件、评测方、日期和可比性
数据、权限、审计、合规、安全与人工兜底
2 条已审核事实 · 全部来自公开来源闸门通过的记录
尚无已审核公开记录的维度:02 能力与任务、03 架构参数、07 部署与接入、08 成本与 TCO、11 业务场景、12 落地与提效。查看资料缺口 ↑
关系只来自年鉴数据库中已登记的归属字段。
工具支持、权重和许可证边界分开判断
README 的 Qwen3 模板说明思考与非思考模式需选择对应 template;本档案使用 qwen3_nothink,不推断其他运行参数。
保留发布日期精度、生命周期节点和原始来源,未公开日期不会被推断。
模型版本 · active
Qwen 官方仓库记录 235B 非思考模式 2507 更新于 2025-07-21 发布。
来源:QwenLM/Qwen3 ↗显示最新核验的 2 条 · 共 2 条已审核事实
Qwen states that Qwen3-235B-A22B-Instruct-2507 was released on 2025-07-21 and Qwen3-235B-A22B-Thinking-2507 was released on 2025-07-25.
Qwen documents 119 languages and dialects for Qwen3. For Qwen3-2507, the maximum assistant message length is 81,920 tokens for thinking models and 16,384 tokens for instruct models; Qwen3-2507 supports 256K context and can be extended to 1M tokens.
1 个版本 · 每项可回到来源与技术解释
高置信度,已绑定来源与证据。
条件:Qwen3-2507 instruct mode
来源:Qwen documentation ↗Qwen documents 119 languages and dialects for Qwen3. For Qwen3-2507, the maximum assistant message length is 81,920 tokens for thinking models and 16,384 tokens for instruct models; Qwen3-2507 supports 256K context and can be extended to 1M tokens.
高置信度,已绑定来源与证据。
条件:Official Qwen3 README News entry.
来源:QwenLM/Qwen3 ↗Qwen states that Qwen3-235B-A22B-Instruct-2507 was released on 2025-07-21 and Qwen3-235B-A22B-Thinking-2507 was released on 2025-07-25.
部分可比 · 测试集:AIME25 · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
引用文本与接口返回都只包含已审核公开事实
阿里千问 / Qwen Qwen3-235B-A22B-Instruct-2507|最新版本 Qwen3-235B-A22B-Instruct-2507|已审核事实 2 条|公开来源 2 个|数据截止 2026-09-17|AI 年鉴|https://ainianjian.com/models/model.alibaba.qwen.qwen3-235b-a22b-instruct-2507
公开接口返回该模型的版本、已审核事实 2 条及其来源信息,可直接用于脚本或对比。
https://api.ainianjian.com/api/models/model.alibaba.qwen.qwen3-235b-a22b-instruct-2507 ↗curl -s 'https://api.ainianjian.com/api/models/model.alibaba.qwen.qwen3-235b-a22b-instruct-2507'
技术规格使用当前版本已审核且来源允许公开采集的事实;模型级开放信息与跨版本地区记录分别标注范围。请按每项适用范围核对,未登记字段显示「未公开」,不使用估算值填充。
部分可比 · 测试集:ARC-AGI · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:Aider-Polyglot · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:Arena-Hard v2 · 条件:Win rate evaluated by GPT-4.1; prompt and harness details are not specified in the table. · 评测方:GPT-4.1 · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:BFCL-v3 · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:CSimpleQA · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:Creative Writing v3 · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:GPQA · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:HMMT25 · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:IFEval · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:INCLUDE · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:LiveBench 20241125 · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:LiveCodeBench v6 (25.02-25.05) · 条件:Official Qwen3-2507 model-card comparison table; benchmark window is part of the test-set name; other harness details are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:MMLU-Pro · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:MMLU-ProX · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:MMLU-Redux · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:MultiIF · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:MultiPL-E · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:PolyMATH · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:RULER 1M · 条件:Full Attention; Dual Chunk Attention enabled; 260 samples per length across 13 sub-tasks (20 samples per sub-task); per-length scores are preserved in evidence. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:RULER 1M · 条件:Sparse Attention; Dual Chunk Attention enabled; 260 samples per length across 13 sub-tasks (20 samples per sub-task); per-length scores are preserved in evidence. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:SimpleQA · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:SuperGPQA · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:TAU1-Airline · 条件:Results generated using GPT-4o-20241120 because native GPT-4o-0327 function calling was unavailable; other prompt and harness details are not specified. · 评测方:GPT-4o-20241120 · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:TAU1-Retail · 条件:Results generated using GPT-4o-20241120 because native GPT-4o-0327 function calling was unavailable; other prompt and harness details are not specified. · 评测方:GPT-4o-20241120 · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:TAU2-Airline · 条件:Results generated using GPT-4o-20241120 because native GPT-4o-0327 function calling was unavailable; other prompt and harness details are not specified. · 评测方:GPT-4o-20241120 · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:TAU2-Retail · 条件:Results generated using GPT-4o-20241120 because native GPT-4o-0327 function calling was unavailable; other prompt and harness details are not specified. · 评测方:GPT-4o-20241120 · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:TAU2-Telecom · 条件:Results generated using GPT-4o-20241120 because native GPT-4o-0327 function calling was unavailable; other prompt and harness details are not specified. · 评测方:GPT-4o-20241120 · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:WritingBench · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗部分可比 · 测试集:ZebraLogic · 条件:Official Qwen3-2507 model-card comparison table; shot count, prompt template, sampling configuration, harness, and evaluation date are not specified. · 评测方:Qwen Team · 日期:未公开
来源:Qwen3-235B-A22B-Instruct-2507 model card ↗