Qwen3.7-Max

36 atoms · 跨 10 天 · 首见 2026-05-21 · 最近 2026-06-26

三色: 🟦 fact 32 · 🟥 take 4 · stance ▲10/▼2/◆21

叙事状态: 🍂 fading (退潮) · 💤 沉寂 · 策展近14d 0 atom

状态只算低频策展源 (群/卖方); X firehose 仅作背景音量

来源: X 36

时态: fresh:35 · stale:1

标签: 好数字:19 · 好信源:12 · 好观点:5 · 好思考:2

🟨 AI 综合 · junior analyst 概览

🟦 客观事实 (facts) (30)

• <a id="atom-91023fb9442a006c"></a>🟦 2026-05-26 X·@poezhao0605 · scored 1541, placing it above GPT-5.5, Gemini-3.5-Flash, GLM-5.1, and Kimi-K2.6 · 竞争格局/模型评分
  Qwen3.7-Max scored 1541, placing it above GPT-5.5, Gemini-3.5-Flash, GLM-5.1, and Kimi-K2.6
  tags: 好数字·好信源 · → daily
  📷 原图
• <a id="atom-f567dbd3cd61e483"></a>🟦 2026-06-26 X·@zerohedge · coding index score · 技术路线/竞争格局
  Qwen3.7-Max代码指数: 50
  tags: 好数字 · → daily
  value: qty=50 · unit=index score
  [图: UBS与Artificial Analysis发布的前沿大模型代码能力与视频生成能力对比图表 — Claude Fable 5代码指数: 62; Qwen3.7-Max代码指数: 50; Seedance 2.0 720p视频生成评分: 1216; HappyHorse-1.0视频生成评分: 1213; Sora2 (December)视频生成评分:
  📷 原图
• <a id="atom-5bb16b08fdfbad38"></a>🟦 2026-05-25 X·@Alibaba_Qwen · implicit caching launch · 产品发布
  ✅Implicit caching is now live on Qwen3.7-Max — kicks in automatically, no setup needed.
  tags: 好数字 · → daily
  value: date=2026-05-25
• <a id="atom-290d0993531e4082"></a>🟦 2026-05-21 X·@ChujieZheng · AA评分表现 · 模型发布/技术路线
  Its top-tier AA score confirms the resulting general and agentic capabilities.
  tags: 好数字 · → daily
  value: qty=top-tier AA score
• <a id="atom-14da7e9f9f67ccc4"></a>🟦 2026-06-18 X·@JustinWaugh · pencil puzzle solve rate · 模型性能
  Qwen3.7-Max currently at 40% per solve
  tags: 好数字 · → daily
  value: qty=40%
  [图: 大语言模型在特定基准测试下的性能表现与成本对比排行榜数据面板 — claude-fable-5@high AGENTIC: 97.6%; gpt-5.5@xhigh AGENTIC: 83.3%; glm-5.2 AGENTIC: 26.7%; glm-5.2 COST/AGENTIC: $20.6620]
  📷 原图
• <a id="atom-6cf3c6e6d3a2c920"></a>🟦 2026-06-18 X·@JustinWaugh · pencil puzzle solve cost · 成本
  Qwen3.7-Max $8.8 per solve
  tags: 好数字 · → daily
  value: qty=$8.8
  [图: 大语言模型在特定基准测试下的性能表现与成本对比排行榜数据面板 — claude-fable-5@high AGENTIC: 97.6%; gpt-5.5@xhigh AGENTIC: 83.3%; glm-5.2 AGENTIC: 26.7%; glm-5.2 COST/AGENTIC: $20.6620]
  📷 原图
• <a id="atom-db46fc8cc020cdc2"></a>🟦 2026-06-01 X·@karminski3 · 后端测试得分 · 模型性能
  后端测试得分6947分
  tags: 好数字 · → daily
  value: qty=6947分 · date=
• <a id="atom-ac0123e395487990"></a>🟦 2026-05-27 X·@Alisvolatprop12 · 参数规模推测 · 供给产能
  보통 max모델은 파라메터가 오픈 웨이트와 달리 1T 내외로 알려져 있습니다
  tags: 好数字 · → daily
  value: qty=1T
• <a id="atom-7ed7abaed69784d2"></a>🟦 2026-05-24 X·@hsu_steve · 性能测试结果 · 性能测试
  The resulting custom kernel achieved a 10x speedup over the official reference code in a 35-hour stress test
  tags: 好数字 · → daily
  value: qty=10x speedup
  📷 原图
• <a id="atom-5277706b4d15488d"></a>🟦 2026-05-22 X·@ZhihuFrontier · 推理性能提升幅度 · 性能提升
  Qwen3.7-Max improves by over 30% compared with 3.6-Preview
  tags: 好数字 · → daily
  value: qty=+30% · date=vs Qwen3.6-Max-Preview
  📷 原图
• <a id="atom-691119d25cf627ef"></a>🟦 2026-05-22 X·@ZhihuFrontier · 每次推理token消耗量 · 推理效率
  Average token usage reaches about 44K
  tags: 好数字 · → daily
  value: qty=~44K · date=当前版本
  📷 原图
• <a id="atom-43a15313b0196f60"></a>🟦 2026-05-22 X·@ZhihuFrontier · token使用量增幅 · 推理效率
  Token usage +50%
  tags: 好数字 · → daily
  value: qty=+50% · date=vs 前代
  📷 原图
• <a id="atom-5c88e240e0d4c4d3"></a>🟦 2026-05-22 X·@ZhihuFrontier · 价格降幅 · 定价
  price drops to 60% of before
  tags: 好数字 · → daily
  value: qty=-40% · date=当前定价
  📷 原图
• <a id="atom-2c29521eb1a86f96"></a>🟦 2026-05-22 X·@ZhihuFrontier · 上下文可靠性排名 · 模型评测
  Among Chinese models, it ranks No.1 in context reliability, second only to GPT-5.5 overall.
  tags: 好数字 · → daily
  value: qty=第二 · date=当前
  📷 原图
• <a id="atom-f5fc670d920cbbc2"></a>🟦 2026-05-22 X·@ZhihuFrontier · 与顶级北美模型的中位差距 · 模型评测
  its median gap to the “big three” is only 4.6%.
  tags: 好数字 · → daily
  value: qty=4.6% · date=当前版本
  📷 原图
• <a id="atom-f1b49a8ee35c4e41"></a>🟦 2026-05-22 X·@ZhihuFrontier · 迭代速度 · 产品发布
  Since 2026, it has almost shipped one update per month.
  tags: 好数字 · → daily
  value: qty=每月一次更新 · date=2026年至今
  📷 原图
• <a id="atom-bc5f0d713d4692e4"></a>🟦 2026-05-22 X·@ZhihuFrontier · 在复杂任务中成本高于DSV4 · 推理效率
  Qwen3.7-Max costs more than DSV4 max mode, and over 2× more than efficient models like GPT-5.5 or Opus 4.6 at similar scores.
  tags: 好数字 · → daily
  value: qty=高于DSV4 max模式 · date=当前
  📷 原图
• <a id="atom-98a23ac47487f12c"></a>🟦 2026-05-21 X·@AlibabaGroup · 模型发布 · 模型发布
  Qwen3.7-Max is live! Introducing the latest proprietary model, built for advanced agentic coding, complex reasoning, and long-horizon execution.
  tags: 好信源 · → daily
  📷 原图
• <a id="atom-ec281d0dda14f012"></a>🟦 2026-05-21 X·@Alibaba_Qwen · 连续自主运行时长 · 技术能力
  35 hours straight on a kernel optimization task — 1,000+ tool calls, zero hand-holding.
  tags: 好数字 · → daily
  value: qty=35 hours · context=kernel optimization task
  📷 原图
• <a id="atom-e306d88a5557354f"></a>🟦 2026-05-31 X·@ZhihuFrontier · 表现显著提升 · 模型发布/竞争格局
  Qwen3.7-Max delivered a significant jump over 3.6
  tags: 好信源 · → daily
• <a id="atom-1f7d621c5bfca965"></a>🟦 2026-05-27 X·@winneravgwin · open-weight 属性 · 技术路线
  핵심은 Qwen3.7-Max가 open-weight라는 점
  tags: 好信源 · → daily
  value: direction=positive
  📷 原图
• <a id="atom-071df9d8c2d0c02a"></a>🟦 2026-05-27 X·@Alisvolatprop12 · 模型类型 · 技术路线
  qwen3.7-max는 오픈 웨이트 모델이 아니라 클라우드/api 모델입니다
  tags: 好信源 · → daily
  value: direction=negative
• <a id="atom-e12dd02e841ad5c6"></a>🟦 2026-05-26 X·@poezhao0605 · designed as an agent foundation model with long-horizon task execution · 技术路线/产品发布
  Qwen3.7-Max is designed as an agent foundation model. Long-horizon task execution and tool calling are core to its architecture.
  tags: 好信源 · → daily
  📷 原图
• <a id="atom-dbfbd5e9b5cd7f76"></a>🟦 2026-05-21 X·@ChujieZheng · RL训练投入计算量 · 模型训练/技术路线
  For Qwen3.7-Max, we have invested far more compute into RL training than ever before.
  tags: 好观点 · → daily
  value: direction=increased
• <a id="atom-1aac318386b70943"></a>🟦 2026-05-21 X·@Alibaba_Qwen · API已上线阿里云Model Studio · 产品发布
  API's up on Alibaba Model Studio.
  tags: 好信源 · → daily
  📷 原图
• <a id="atom-3901ab29561f0a5d"></a>🟦 2026-05-21 X·@Ali_TongyiLab · is built for the agent era · 模型发布
  Meet Qwen3.7-Max, our latest proprietary model built specifically for the agent era.
  tags: 好信源 · → daily
• <a id="atom-9b8279f5862eb1cf"></a>🟦 2026-05-21 X·@Ali_TongyiLab · excels at Frontier Coding Agent · 能力描述
  Frontier Coding Agent: Excels from frontend prototyping to complex software engineering.
  tags: 好信源 · → daily
• <a id="atom-fba178025e5b3dfa"></a>🟦 2026-05-21 X·@Ali_TongyiLab · serves as Cowork Productivity Assistant · 能力描述
  Cowork Productivity Assistant: Reliable office assistant and workflow automation via MCP and multi-agent orchestration.
  tags: 好信源 · → daily
• <a id="atom-0c24a3379532d4a3"></a>🟦 2026-05-21 X·@Ali_TongyiLab · supports Long-Horizon Autonomous Execution · 能力描述
  Long-Horizon Autonomous Execution: Built for sustained, autonomous execution across long-horizon tasks.
  tags: 好信源 · → daily
• <a id="atom-4175d49fa9fdb243"></a>🟦 2026-05-21 X·@Ali_TongyiLab · enables Cross-Harness Generalization · 能力描述
  Cross-Harness Generalization: Consistent performance across diverse agent frameworks.
  tags: 好信源 · → daily

🟥 多头 takes (bullish) (2)

• <a id="atom-d67abdafc606422f"></a>🟥 2026-05-21 X·@ChujieZheng ⭐ · 强化学习扩展计划 · 技术路线/产品发布
  We will firmly push forward RL scaling to build more powerful Qwen models.
  tags: 好观点·好思考 · → daily
  value: direction=increased
• <a id="atom-bc893300dd6ff945"></a>🟥 2026-05-22 X·@ZhihuFrontier · 在推理稳定性、上下文可靠性和实用编码方面进步显著 · 模型评测
  Qwen3.7-Max takes another real step forward, especially in reasoning stability, context reliability, and practical coding usability.
  tags: 好观点 · → daily
  📷 原图

🟥 空头 takes (bearish) (1)

• <a id="atom-24af65873990fb89"></a>🟥 2026-05-21 X·@scaling01 [老旧] · 定价 · 价格动态
  pretty expensive at $2.5 / $7.5
  tags: 好数字 · → daily
  value: qty=$2.5 / $7.5 · date=2026-05-21

🟥 中性 takes (neutral) (1)

• <a id="atom-8bf3427101e2b2ac"></a>🟥 2026-05-21 X·@scaling01 · 模型参数量范围推测 · 技术路线
  but I would guess it's also in the 1T-2T range
  tags: 好思考 · → daily
  value: qty=1T-2T · date=2026-05-21

⏱ 时间轴 (近 20)

• 🟦 2026-06-26 · fact · coding index score · → daily
• 🟦 2026-06-18 · fact · pencil puzzle solve rate · → daily
• 🟦 2026-06-18 · fact · pencil puzzle solve cost · → daily
• 🟦 2026-06-01 · fact · 后端测试得分 · → daily
• 🟦 2026-06-01 · fact · 实现了IVF-PQ+ADC索引方案 · → daily
• 🟦 2026-05-31 · fact · 表现显著提升 · → daily
• 🟦 2026-05-27 · fact · open-weight 属性 · → daily
• 🟦 2026-05-27 · refute · 模型类型 · → daily
• 🟦 2026-05-27 · fact · 参数规模推测 · → daily
• 🟦 2026-05-26 · fact · designed as an agent foundation model with long-horizon task · → daily
• 🟦 2026-05-26 · fact · scored 1541, placing it above GPT-5.5, Gemini-3.5-Flash, GLM · → daily
• 🟦 2026-05-25 · fact · implicit caching launch · → daily
• 🟦 2026-05-24 · fact · 性能测试结果 · → daily
• 🟦 2026-05-22 · fact · 推理性能提升幅度 · → daily
• 🟦 2026-05-22 · fact · 每次推理token消耗量 · → daily
• 🟦 2026-05-22 · fact · token使用量增幅 · → daily
• 🟦 2026-05-22 · fact · 价格降幅 · → daily
• 🟦 2026-05-22 · fact · 上下文可靠性排名 · → daily
• 🟦 2026-05-22 · fact · 与顶级北美模型的中位差距 · → daily
• 🟦 2026-05-22 · fact · 迭代速度 · → daily

← 实体目录 · 系统日志