以下部分引用 AI 总结，现阶段 AI 仍然有幻觉。请以内容的原文为准。

数据新鲜度群聊 07-14 ✓卖方 07-14 ✓Wrap 断档15天 ✕News 07-15 ✓X 断档12天 ✕页面生成 07-15

Qwen

100 atoms · 跨 45 天 · 首见 2026-04-16 · 最近 2026-06-29

三色: 🟦 fact 60 · 🟥 take 40 · stance ▲28/▼12/◆29

叙事状态: 🍂 fading (退潮) · 💤 沉寂 · 策展近14d 0 atom · 🐦 X 近14d 1

状态只算低频策展源 (群/卖方); X firehose 仅作背景音量

来源: 群 1 · X 99

时态: fresh:90 · aging:8 · stale:2

标签: 好信源:37 · 好观点:34 · 好数字:26 · 好思考:6 · 好问题:3

🟨 AI 综合 · junior analyst 概览

展开 AI 综合 (灰色 · 非市场结论 · 点击数字溯源到原 atom)

model: deepseek-chat · 2026-07-12 · 默认折叠

Qwen 在多模态领域保持领先，且密集发布杀手级模型，包括即将推出的 27B 和 35B 新模型 ¹ ¹ ¹。其市场影响力仍占绝大部分份额，并一个接一个地发布世界模型 ¹ ¹。然而，其开放策略正转向更封闭的方向，且本地模型远不如闭源对手 ¹ ¹。Qwen 在写作能力上从未表现出色，成本效率也极其糟糕 ¹ ¹。KV 缓存机制需要手动初始化，这导致了隐藏的冗余推理成本 ¹。尽管定价远低于美国竞品，但受 DeepSeek 降价影响，其市场份额反而受到挤压 ¹ ¹。

🧭 拥挤度 (一人一票): ▲ 12 位作者 (KOL12) vs ▼ 7 位作者 (KOL7) + 群 1 条

⚖️ 多头 12/12 来自KOL

💢 核心分歧 (4 轴)

1. 模型发布节奏 vs 实际能力落地

*topic: 模型发布/技术路线 · 6 bull vs 1 bear*

🟢 bullish 侧:

<a id="atom-71b88e96ab421a91"></a>2026-04-22 X·@zephyr_z9 · 发布密集模型 · 产品发布/模型发布
Qwen is releasing some killer dense models
tags: 好信源 · → daily
<a id="atom-a969d7230a2d7636"></a>2026-04-27 X·@ZhihuFrontier [老化中] · Qwen3.6 Flash shows strong local deployment efficiency and tool-use capability · 模型发布
Qwen｜Qwen3.6 Flash shows strong local deployment efficiency and tool-use capability
tags: 好观点 · → daily
value: date=2026-04-27
📷 原图
<a id="atom-3b66dff3e86067c5"></a>2026-05-18 X·@poezhao0605 · 新旗舰模型发布预告 · 模型发布
Qwen is teasing a major new model drop at the Alibaba Cloud Summit on May 20. Keywords from the official teaser: “more versatile, more powerful, deeper, broader.” Sounds like a flagship update is coming.
tags: 好观点 · → daily
📷 原图
<a id="atom-c428e78601d919ee"></a>2026-05-18 X·@leftcurvedev_ · 新模型发布 · 模型发布
Qwen3.7 family incoming — upcoming new 27B & 35B models
tags: 好思考 · → daily
value: qty=27B & 35B · date=2026
<a id="atom-d6a72fef9306b325"></a>2026-05-25 X·@advait_jayant · 模型出货状态 · 模型发布
deepseek, qwen, kimi, minimax, and as you pointed out even meituan and xiaomi are all shipping their own models!
tags: 好信源 · → daily
value: direction=positive

🔴 bearish 侧:

<a id="atom-9792842169496e07"></a>2026-06-16 X·@IlyasHairline · 模型质量 · 模型发布
is qwen even that good? because no local model except 120b was of any use for me (and still kinda slow)
tags: 好问题 · → daily

展开 13 条中性

<a id="atom-a91f0613c2e903c3"></a>2026-04-23 X·@Xianbao_QIAN · 模型参数量 · 模型发布
1.6T params
tags: 好数字 · → daily
value: qty=1.6T · date=2026-04-24
📷 原图
<a id="atom-d64da2e62bfa0427"></a>2026-05-19 X·@poezhao0605 · next model upgrade for agent workloads · 模型发布
Zhou Jingren on Qwen’s next model upgrade, built specifically for agent workloads.
tags: 好信源 · → daily
📷 原图
<a id="atom-9b03f7ef994a7651"></a>2026-05-31 X·@jenzhuscott · coding benchmark performance · 技术路线/模型发布
Qwen just matched Claude Opus on coding benchmarks.
tags: 好数字 · → daily
<a id="atom-6815d27d5d39f455"></a>2026-06-01 X·@scaling01 · 模型发布 · 模型发布
Qwen just released Qwen 3.7 Plus
tags: 好信源 · → daily
<a id="atom-aae29280d9c11056"></a>2026-06-14 X·@AiBattle_ · Max/Plus 模型发布节奏 · 模型发布
Qwen has also been releasing monthly iterations of their Max / Plus model
tags: 好信源 · → daily
value: qty=monthly
📷 原图

2. 定价优势 vs 成本效率劣势

*topic: 竞争格局/开源/推理成本 · 5 bull vs 6 bear*

🟢 bullish 侧:

<a id="atom-585b428712f7525e"></a>2026-04-22 X·@georgejrjrjr · 30B参数层级免作弊评测表现 · 技术路线/竞争格局
qwen is one of few labs giving goog a run for its money on uncheatable eval in the ~30B tier
tags: 好数字·好观点 · → daily
<a id="atom-ca5cb433912e9889"></a>2026-05-26 X·@natolambert [老旧] · leader status in open models · 竞争格局/模型发布
Qwen is still king
tags: 好观点 · → daily
<a id="atom-e57f93d966f8e3bb"></a>2026-05-29 X·@teortaxesTex · 在多模态领域保持领先 · 技术路线/竞争格局
Qwen isn't giving up its leadership in multimodality.
tags: 好观点 · → daily
<a id="atom-e9b27a3fb85ece2a"></a>2026-06-03 X·@TheAhmadOsman [老旧] · 中国实验室 tier 评级 · 竞争格局
A-tier is Zhipu AI and Qwen for me
tags: 好观点 · → daily
value: qty=A-Tier · date=2026-06-03
<a id="atom-575676c5b354d270"></a>2026-06-16 X·@teortaxesTex · 涉足领域 · 技术路线/竞争格局
This is the Nvidia/GDM turf
tags: 好观点 · → daily
value: direction=Nvidia/GDM turf
📷 原图

🔴 bearish 侧:

<a id="atom-4835894aa9108419"></a>2026-04-25 X·@teortaxesTex [老化中] · Hybrid Linear is inferior to DeepSeek Hybrid Compressed · 竞争格局/技术路线
yeah their Hybrid Linear is just inferior to DeepSeek's Hybrid Compressed
tags: 好观点 · → daily
<a id="atom-b893f2a75cca09f7"></a>2026-04-21 X·@Yuchenj_UW [老化中] · 开放策略方向 · 开源
Qwen is moving in a more closed direction
tags: 好思考 · → daily
value: direction=更封闭
<a id="atom-56ded6b2c79801b3"></a>2026-05-20 X·@zhenbailian · GLM scores alteration allegation · 竞争格局/技术路线
So they altered GLM's official scores? Interesting. It was definitely not as good as the benchmarks suggested.
tags: 好问题 · → daily
<a id="atom-eb0c10270df53042"></a>2026-05-22 X·@manateelazycat · 受 DeepSeek 降价影响 · 竞争格局
反而受到影响的是国内几家，比如Qwen、GLM、MiniMax 等
tags: 好观点 · → daily
value: statement=受到影响
📷 原图
<a id="atom-b1f79439ef0e3c95"></a>2026-06-02 X·@teortaxesTex · 成本效率 · 竞争格局
we also see abysmal cost-efficiency from Qwen
tags: 好观点 · → daily
value: qty=abysmal · direction=cost-efficiency
📷 原图

展开 4 条中性

<a id="atom-c76f5d9f7c5e031f"></a>2026-05-25 X·@Gracemzshao · 利用开放性吸引关注、招募人才、加速生态 · 竞争格局/人才获用
Qwen uses openness as a way to win attention, recruit talent, and accelerate the whole ecosystem
tags: 好观点 · → daily
📷 原图
<a id="atom-97a2528cd3b4a4a7"></a>2026-06-01 X·@Presidentlin · 退出开源模型竞赛 · 竞争格局
Qwen exited the OSS race at the end of the day.
tags: 好信源 · → daily
📷 原图
<a id="atom-ddd0a3759811c88a"></a>2026-06-16 X·@teortaxesTex · 发布模型 · 模型发布/竞争格局
A really massive release from Qwen
tags: 好信源 · → daily
value: date=2026-06-16 · direction=released
📷 原图
<a id="atom-43620d01bcbd3926"></a>2026-05-29 X·@sun_hanchi · 推理成本上限 · 推理成本
Qwen’s inference cost is at most $0.5
tags: 好数字 · → daily
value: qty=$0.5

3. 本地模型 vs 闭源 API 性能差距

*topic: 模型表现/模型性能 · 1 bull vs 3 bear*

🟢 bullish 侧:

<a id="atom-b034c579d26f731f"></a>2026-04-23 X·@georgejrjrjr [老化中] · benchmaxxing helps benchmarks · 模型性能
benchmaxxing helps benchmarks -> training on every conceivable future test-set as they do is going to show up as an edge for Qwen, too, in translating base model performance to downstream evals.
tags: 好观点 · → daily

🔴 bearish 侧:

<a id="atom-c68f87032a342286"></a>2026-06-28 群 · 本地模型vs闭源 · 模型表现
Tae Kim实测：Qwen本地模型远不如OpenAI/Anthropic
tags: 好信源 · → daily
<a id="atom-9cff1298a0047c37"></a>2026-06-14 X·@_xjdr · 模型工作可用性 · 模型性能_
i've just never had real success with their models for real work
tags: 好观点 · → daily
<a id="atom-99f7ee0e9c1a0cb2"></a>2026-06-23 X·@teortaxesTex · 写作capability · 模型性能
qwen has never been good at writing, in any language
tags: 好观点 · → daily
value: direction=not good

4. 技术路线领先性 vs 被对手压缩替代

*topic: 技术路线/竞争格局 · 10 bull vs 1 bear*

🟢 bullish 侧:

<a id="atom-5305c8f298fe9e58"></a>2026-04-22 X·@georgejrjrjr · 模型向STEM/代码域泛化能力 · 模型发布/技术路线
this turns out to generalize surprisingly well within the domains they care about (STEM / code)
tags: 好观点 · → daily
<a id="atom-80bcfade9349916d"></a>2026-04-25 X·@teortaxesTex [老化中] · model performance · 竞争格局
Total Qwen victory… Whale is beached and washed.
tags: 好观点 · → daily
value: qty=victory · direction=up
📷 原图
<a id="atom-33e80d2b91071e23"></a>2026-05-05 X·@TheAhmadOsman [老化中] · Qwen 2.5 为最具有重要意义开放权重模型之一 · 技术路线/模型发布
The 3 most consequential open weight releases - Qwen 2.5
tags: 好观点 · → daily
<a id="atom-0075702ca59cc424"></a>2026-05-20 X·@teortaxesTex · benchmark performance progress from 3.6-Max to 3.7-Max · 模型发布/技术路线
Composite table for the four benchmarks where Qwen has shown both 3.6-Max (Preview) and 3.7-Max. The progress is not exactly dramatic, but it is significant for 1 month.
tags: 好数字 · → daily
value: date=1 month
📷 原图
<a id="atom-d9b42070c9208259"></a>2026-05-27 X·@LottoLabs · 3.7 版本对 70B 模型能力影响 · 模型发布/技术路线
Qwen 3.7 could make 70b models great again
tags: 好观点 · → daily

🔴 bearish 侧:

<a id="atom-b2b730c442d88373"></a>2026-05-30 X·@teortaxesTex · still has token slips in Russian · 技术路线
Big Qwen still has token slips
tags: 好观点 · → daily

展开 4 条中性

<a id="atom-fd16e8f0506adb1b"></a>2026-05-31 X·@ZhihuFrontier · 推出WebWorld · 技术路线
Qwen introduces WebWorld, a browser simulator for training Agents without relying on the real internet, together with WebWorldData trajectories.
tags: 好信源 · → daily
<a id="atom-fe82c0a378c3fdce"></a>2026-06-26 X·@anindyadeeps · 200-300B参数模型效果优秀 · 技术路线/模型发布
you can always train very good models in the 200-300B parameter regime, like how Qwen did
tags: 好观点 · → daily
value: qty=200-300B
<a id="atom-c8143d73918af7c4"></a>2026-06-26 X·@max_paperclips · run through an abliteration now · 技术路线
I'm already in the habit of running Qwen through an abliteration now (on general principle), other labs will just do that & lay their own training on top
tags: 好思考 · → daily
<a id="atom-7717b62a10e3fd94"></a>2026-06-29 X·@ChinaMacroFacts · 预期在接下来几个月会突破模型能力阈值 · 技术路线
如果接下来几个月，我们能看到qwen、minimax、deepseek逐个突破
tags: 好观点 · → daily

🟦 客观事实 (facts) (30)

<a id="atom-2c51df78ff6fec5c"></a>🟦 2026-06-27 X·@rohanpaul_ai · 定价低于美国竞品 · 定价权/竞争格局
Chinese AI models are up to 50 times cheaper than their American counterparts on a per-token basis. Especially Qwen, DeepSeek and Kimi, which pressures OpenAI and Anthropic pricing.
tags: 好数字·好信源 · → daily
value: direction=up to 50 times cheaper
📷 原图
<a id="atom-a7512dad755aabae"></a>🟦 2026-05-21 X·@testingcatalog · 模型发布 · 模型发布/技术路线
Alibaba released Qwen 3.7 Max, its latest proprietary model for agentic coding. Qwen 3.7 Max scores 56.6 on the Artificial Analysis Intelligence Index
tags: 好信源·好数字 · → daily
value: qty=56.6 · date=2026-05-21
📷 原图
<a id="atom-c406838e51fae067"></a>🟦 2026-05-18 X·@leftcurvedev_ · Arena 排名 · 模型发布
In Text Arena, Qwen3.7 Max Preview ranks #13 overall.
tags: 好数字·好信源 · → daily
value: qty=#13 overall
<a id="atom-585b428712f7525e"></a>🟦 2026-04-22 X·@georgejrjrjr · 30B参数层级免作弊评测表现 · 技术路线/竞争格局
qwen is one of few labs giving goog a run for its money on uncheatable eval in the ~30B tier
tags: 好数字·好观点 · → daily
<a id="atom-c6bf9d82c7b2b9bb"></a>🟦 2026-06-26 X·@max_paperclips · 模型发布 · 模型发布
Qwen are releasing world model after world model
tags: 好数字 · → daily
<a id="atom-7c08cad95a429399"></a>🟦 2026-06-23 X·@AiBattle_ · open-weight release plan · 模型发布
Qwen staff: Absolutely will do
tags: 好信源 · → daily
value: direction=confirmed
📷 原图
<a id="atom-6815d27d5d39f455"></a>🟦 2026-06-01 X·@scaling01 · 模型发布 · 模型发布
Qwen just released Qwen 3.7 Plus
tags: 好信源 · → daily
<a id="atom-9ee3a1b2887665d5"></a>🟦 2026-04-24 X·@Techmeme · 与车企集成 · 合作客户
Alibaba says its Qwen AI models will be integrated into BYD, Volkswagen, and other cars
tags: 好信源 · → daily
value: qty=BYD, Volkswagen, 其他车型
<a id="atom-d56e492f1650c26a"></a>🟦 2026-04-24 X·@Techmeme · 车载功能 · 产品发布
letting users buy food and tickets via voice commands on select models
tags: 好信源 · → daily
value: detail=语音购票、订餐
<a id="atom-c9dd7d3702c6b04f"></a>🟦 2026-06-24 X·@HuggingPapers · trained on 10M+ trajectories · 模型发布
Trained on 10M+ trajectories, the models serve as both scalable simulators and agent foundation models.
tags: 好数字 · → daily
value: qty=10M+ · unit=trajectories
📷 原图
<a id="atom-3a672694bc1a8173"></a>🟦 2026-06-23 X·@HuggingPapers · 数据集包含2170轮评估 · 基准测试
2,170 turns scoring how well AI simulates environment states on format, factuality, consistency, realism, and quality across 7 agent domains
tags: 好数字 · → daily
value: qty=2,170 turns
<a id="atom-242ea999256a8950"></a>🟦 2026-06-07 X·@LottoLabs · top downloaded models all under 9b · 模型发布/订单需求
top downloaded qwen models All < 9b
tags: 好数字 · → daily
📷 原图
<a id="atom-b37189f31e68ba6f"></a>🟦 2026-06-03 X·@poezhao0605 · MAU 变化 · 用户规模
Alibaba's Qwen grew 6% to 234 million MAU.
tags: 好数字 · → daily
value: qty=+6% · date=2026-06-03
📷 原图
<a id="atom-9b03f7ef994a7651"></a>🟦 2026-05-31 X·@jenzhuscott · coding benchmark performance · 技术路线/模型发布
Qwen just matched Claude Opus on coding benchmarks.
tags: 好数字 · → daily
<a id="atom-43620d01bcbd3926"></a>🟦 2026-05-29 X·@sun_hanchi · 推理成本上限 · 推理成本
Qwen’s inference cost is at most $0.5
tags: 好数字 · → daily
value: qty=$0.5
<a id="atom-4994603b13d55e5b"></a>🟦 2026-05-28 X·@lewtun · 所用RL训练环境规模 · 技术路线_
Today's best coding models from Qwen, DeepSeek, Minimax etc are trained on 1000s of concurrent RL environments
tags: 好数字 · → daily
value: qty=1000s of concurrent RL environments
<a id="atom-3080b0f1655cd33f"></a>🟦 2026-05-27 X·@Alisvolatprop12 · 公开开放权重模型版本 · 供给产能
오픈웨이트 qwen은 3.6(35B;MoE, 27B:Dense)까지만 공개되어 있습니다
tags: 好数字 · → daily
value: qty=3.6(35B;MoE, 27B:Dense)
<a id="atom-0075702ca59cc424"></a>🟦 2026-05-20 X·@teortaxesTex · benchmark performance progress from 3.6-Max to 3.7-Max · 模型发布/技术路线
Composite table for the four benchmarks where Qwen has shown both 3.6-Max (Preview) and 3.7-Max. The progress is not exactly dramatic, but it is significant for 1 month.
tags: 好数字 · → daily
value: date=1 month
📷 原图
<a id="atom-98be32ef90ab8598"></a>🟦 2026-05-14 X·@Alex_Tsico · 价格 · 价格动态
Qwen à 0,50$
tags: 好数字 · → daily
value: qty=$0.50 · unit=每百万 tokens
<a id="atom-c55f887fde7c6ec3"></a>🟦 2026-04-28 X·@arankomatsuzaki · non-English token tax · 模型效率/token效率
Gemini and Qwen provide least non-English tax
tags: 好数字 · → daily
value: direction=least
📷 原图
<a id="atom-771b4162206ec812"></a>🟦 2026-04-27 X·@poezhao0605 · 市场份额 · 市场份额
Qwen: 13% share of China's AI app iOS downloads in March 2026
tags: 好数字 · → daily
value: qty=13% · date=2026-03
[图: 2026年3月中国本土AI应用iOS下载量份额排行条形图 — Doubao: 20%; Jimeng AI: 17%; Qwen: 13%; Yuanbao: 12%; Ant Afu: 11%]
📷 原图
<a id="atom-75bb2608974f3ab0"></a>🟦 2026-04-25 X·@teortaxesTex · token效率 · 模型效率
Qwen uses 40% fewer tokens
tags: 好数字 · → daily
value: qty=40% fewer tokens · direction=优于
📷 原图
<a id="atom-ba63baeb4ce79dee"></a>🟦 2026-04-25 X·@teortaxesTex · 推理时间效率 · 模型效率
Qwen uses 10% less time
tags: 好数字 · → daily
value: qty=10% less time · direction=优于
📷 原图
<a id="atom-18552bf95e73b167"></a>🟦 2026-04-25 X·@teortaxesTex · 推理成本 · 模型成本
Qwen uses 3.6x more money
tags: 好数字 · → daily
value: qty=3.6x more money
📷 原图
<a id="atom-a91f0613c2e903c3"></a>🟦 2026-04-23 X·@Xianbao_QIAN · 模型参数量 · 模型发布
1.6T params
tags: 好数字 · → daily
value: qty=1.6T · date=2026-04-24
📷 原图
<a id="atom-aee03ceeb6ee9c84"></a>🟦 2026-04-23 X·@Xianbao_QIAN · 模型预训练数据量 · 模型发布
33T pretrain data
tags: 好数字 · → daily
value: qty=33T · date=2026-04-24
📷 原图
<a id="atom-ecbaa24ea8add6b5"></a>🟦 2026-04-23 X·@Xianbao_QIAN · 模型上下文长度 · 模型发布
Context length 1M
tags: 好数字 · → daily
value: qty=1M · date=2026-04-24
📷 原图
<a id="atom-b0234adf08f6f64d"></a>🟦 2026-06-25 X·@Xianbao_QIAN · 招聘多模态智能体实习生 · 人事变动
Qwen team is hiring multimodal agent intern.
tags: 好信源 · → daily
<a id="atom-1e30e0ceca9b6ce5"></a>🟦 2026-06-24 X·@HuggingPapers · 发布了 AgentWorld · 模型发布
Qwen just released AgentWorld on Hugging Face
tags: 好信源 · → daily
📷 原图
<a id="atom-1586c604d0f96ca4"></a>🟦 2026-06-24 X·@HuggingPapers · released native language world models that simulate 7 agentic domains · 模型发布
Qwen-AgentWorld: Alibaba's Qwen team released native language world models that simulate 7 agentic domains via long chain-of-thought reasoning.
tags: 好信源 · → daily
value: qty=7 · unit=domains
📷 原图

🟥 多头 takes (bullish) (18)

<a id="atom-9d0f585b0d76f622"></a>🟥 2026-06-14 X·@Presidentlin · 领先程度 · 竞争格局
Ok ahead of Qwen (0) but behind Mistral (has 13)
tags: 好数字·好观点 · → daily
value: qty=0
<a id="atom-71b88e96ab421a91"></a>🟥 2026-04-22 X·@zephyr_z9 · 发布密集模型 · 产品发布/模型发布
Qwen is releasing some killer dense models
tags: 好信源 · → daily
<a id="atom-e57f93d966f8e3bb"></a>🟥 2026-05-29 X·@teortaxesTex · 在多模态领域保持领先 · 技术路线/竞争格局
Qwen isn't giving up its leadership in multimodality.
tags: 好观点 · → daily
<a id="atom-c428e78601d919ee"></a>🟥 2026-05-18 X·@leftcurvedev_ · 新模型发布 · 模型发布
Qwen3.7 family incoming — upcoming new 27B & 35B models
tags: 好思考 · → daily
value: qty=27B & 35B · date=2026
<a id="atom-0c64df1bf91dd30f"></a>🟥 2026-06-25 X·@xeophon · has the vast majority of market share · 市场份额
Overall impact is still dominated by Qwen, which has the vast majority of market share
tags: 好观点 · → daily
<a id="atom-2d8a573de1708e33"></a>🟥 2026-06-25 X·@max_paperclips · model release frequency · 模型发布/技术路线
Qwen releasing world model after world model
tags: 好观点 · → daily
<a id="atom-a5f34d103cd6cbaa"></a>🟥 2026-06-22 X·@teortaxesTex · trajectory · 技术路线
Qwen is on a good trajectory too
tags: 好观点 · → daily
value: direction=good
<a id="atom-575676c5b354d270"></a>🟥 2026-06-16 X·@teortaxesTex · 涉足领域 · 技术路线/竞争格局
This is the Nvidia/GDM turf
tags: 好观点 · → daily
value: direction=Nvidia/GDM turf
📷 原图
<a id="atom-ff199358b94f94fd"></a>🟥 2026-06-13 X·@LottoLabs · 模型能力级别达到27B并可在移动端运行 · 模型发布/技术路线
We’re going to get qwen 27b level models on mobile within the next year
tags: 好观点 · → daily
value: qty=27B · date=within the next year
<a id="atom-e9b27a3fb85ece2a"></a>🟥 2026-06-03 X·@TheAhmadOsman [老旧] · 中国实验室 tier 评级 · 竞争格局
A-tier is Zhipu AI and Qwen for me
tags: 好观点 · → daily
value: qty=A-Tier · date=2026-06-03
<a id="atom-d9b42070c9208259"></a>🟥 2026-05-27 X·@LottoLabs · 3.7 版本对 70B 模型能力影响 · 模型发布/技术路线
Qwen 3.7 could make 70b models great again
tags: 好观点 · → daily
<a id="atom-ca5cb433912e9889"></a>🟥 2026-05-26 X·@natolambert [老旧] · leader status in open models · 竞争格局/模型发布
Qwen is still king
tags: 好观点 · → daily
<a id="atom-69cc235d8181b8ee"></a>🟥 2026-05-25 X·@TheAhmadOsman · Small Specialized models equal out · 模型小型化
As we get closer and closer to Small & Specialized models, it will all equal out (e.g. Qwen 3.5 27B)
tags: 好观点 · → daily
📷 原图
<a id="atom-3b66dff3e86067c5"></a>🟥 2026-05-18 X·@poezhao0605 · 新旗舰模型发布预告 · 模型发布
Qwen is teasing a major new model drop at the Alibaba Cloud Summit on May 20. Keywords from the official teaser: “more versatile, more powerful, deeper, broader.” Sounds like a flagship update is coming.
tags: 好观点 · → daily
📷 原图
<a id="atom-33e80d2b91071e23"></a>🟥 2026-05-05 X·@TheAhmadOsman [老化中] · Qwen 2.5 为最具有重要意义开放权重模型之一 · 技术路线/模型发布
The 3 most consequential open weight releases - Qwen 2.5
tags: 好观点 · → daily
<a id="atom-a969d7230a2d7636"></a>🟥 2026-04-27 X·@ZhihuFrontier [老化中] · Qwen3.6 Flash shows strong local deployment efficiency and tool-use capability · 模型发布
Qwen｜Qwen3.6 Flash shows strong local deployment efficiency and tool-use capability
tags: 好观点 · → daily
value: date=2026-04-27
📷 原图
<a id="atom-80bcfade9349916d"></a>🟥 2026-04-25 X·@teortaxesTex [老化中] · model performance · 竞争格局
Total Qwen victory… Whale is beached and washed.
tags: 好观点 · → daily
value: qty=victory · direction=up
📷 原图
<a id="atom-b034c579d26f731f"></a>🟥 2026-04-23 X·@georgejrjrjr [老化中] · benchmaxxing helps benchmarks · 模型性能
benchmaxxing helps benchmarks -> training on every conceivable future test-set as they do is going to show up as an edge for Qwen, too, in translating base model performance to downstream evals.
tags: 好观点 · → daily

🟥 空头 takes (bearish) (11)

<a id="atom-d64386cb8085495d"></a>🟥 2026-05-09 X·@ZhihuFrontier · KV Cache mechanism · 推理成本/缓存架构
Qwen / Seed / Mimo adopt no automatic KV Cache mechanism — require manual cache initialization and additional charging. No natural TTL retention, leading to hidden redundant inference costs for regular users.
tags: 好信源·好思考 · → daily
[图: 大语言模型API Token使用量与缓存命中情况的数据面板图表 — 总Token数: 50,949,110; 2026-04-28单日Token数: 21,213,830; 输入(命中缓存): 20,526,208; 输入(未命中缓存): 443,048; 输出: 244,574]
📷 原图
<a id="atom-c68f87032a342286"></a>🟥 2026-06-28 群 · 本地模型vs闭源 · 模型表现
Tae Kim实测：Qwen本地模型远不如OpenAI/Anthropic
tags: 好信源 · → daily
<a id="atom-b893f2a75cca09f7"></a>🟥 2026-04-21 X·@Yuchenj_UW [老化中] · 开放策略方向 · 开源
Qwen is moving in a more closed direction
tags: 好思考 · → daily
value: direction=更封闭
<a id="atom-99f7ee0e9c1a0cb2"></a>🟥 2026-06-23 X·@teortaxesTex · 写作capability · 模型性能
qwen has never been good at writing, in any language
tags: 好观点 · → daily
value: direction=not good
<a id="atom-9cff1298a0047c37"></a>🟥 2026-06-14 X·@_xjdr · 模型工作可用性 · 模型性能_
i've just never had real success with their models for real work
tags: 好观点 · → daily
<a id="atom-b1f79439ef0e3c95"></a>🟥 2026-06-02 X·@teortaxesTex · 成本效率 · 竞争格局
we also see abysmal cost-efficiency from Qwen
tags: 好观点 · → daily
value: qty=abysmal · direction=cost-efficiency
📷 原图
<a id="atom-eb0c10270df53042"></a>🟥 2026-05-22 X·@manateelazycat · 受 DeepSeek 降价影响 · 竞争格局
反而受到影响的是国内几家，比如Qwen、GLM、MiniMax 等
tags: 好观点 · → daily
value: statement=受到影响
📷 原图
<a id="atom-819cb3d7c9306010"></a>🟥 2026-05-12 X·@teortaxesTex [老化中] · small model training cost · 模型训练成本
Small Awen models are even more expensive to train than I imagined
tags: 好观点 · → daily
value: qty=more expensive · direction=higher
<a id="atom-4835894aa9108419"></a>🟥 2026-04-25 X·@teortaxesTex [老化中] · Hybrid Linear is inferior to DeepSeek Hybrid Compressed · 竞争格局/技术路线
yeah their Hybrid Linear is just inferior to DeepSeek's Hybrid Compressed
tags: 好观点 · → daily
<a id="atom-9792842169496e07"></a>🟥 2026-06-16 X·@IlyasHairline · 模型质量 · 模型发布
is qwen even that good? because no local model except 120b was of any use for me (and still kinda slow)
tags: 好问题 · → daily
<a id="atom-56ded6b2c79801b3"></a>🟥 2026-05-20 X·@zhenbailian · GLM scores alteration allegation · 竞争格局/技术路线
So they altered GLM's official scores? Interesting. It was definitely not as good as the benchmarks suggested.
tags: 好问题 · → daily

🟥 中性 takes (neutral) (7)

<a id="atom-97a2528cd3b4a4a7"></a>🟥 2026-06-01 X·@Presidentlin · 退出开源模型竞赛 · 竞争格局
Qwen exited the OSS race at the end of the day.
tags: 好信源 · → daily
📷 原图
<a id="atom-c8143d73918af7c4"></a>🟥 2026-06-26 X·@max_paperclips · run through an abliteration now · 技术路线
I'm already in the habit of running Qwen through an abliteration now (on general principle), other labs will just do that & lay their own training on top
tags: 好思考 · → daily
<a id="atom-437f991f07794f1f"></a>🟥 2026-05-02 X·@Presidentlin [老化中] · drip feeding release strategy · 产品发布
Qwen, in the early days, picked up a nice little trick of drip feeding your releases, instead of doing a big bang, everything all at once.
tags: 好思考 · → daily
<a id="atom-7717b62a10e3fd94"></a>🟥 2026-06-29 X·@ChinaMacroFacts · 预期在接下来几个月会突破模型能力阈值 · 技术路线
如果接下来几个月，我们能看到qwen、minimax、deepseek逐个突破
tags: 好观点 · → daily
<a id="atom-fe82c0a378c3fdce"></a>🟥 2026-06-26 X·@anindyadeeps · 200-300B参数模型效果优秀 · 技术路线/模型发布
you can always train very good models in the 200-300B parameter regime, like how Qwen did
tags: 好观点 · → daily
value: qty=200-300B
<a id="atom-c76f5d9f7c5e031f"></a>🟥 2026-05-25 X·@Gracemzshao · 利用开放性吸引关注、招募人才、加速生态 · 竞争格局/人才获用
Qwen uses openness as a way to win attention, recruit talent, and accelerate the whole ecosystem
tags: 好观点 · → daily
📷 原图
<a id="atom-e24dff1bbe52ad3b"></a>🟥 2026-06-14 X·@AiBattle_ · 新模型可能即将发布 · 模型发布
We might see a new release from them as well
tags: 好问题 · → daily
value: date=June 2026
📷 原图

🟦 新闻流 (squawk · 2)

展开新闻流 (FirstSquawk / financialjuice / wallstengine / DeItaone — 快讯, 非原创 take)

<a id="atom-5da640984d999200"></a>🟦 2026-06-15 X·@wallstengine · 收益率 · 仓位情绪
Chinese AI model Qwen is down 35.6%
tags: 好数字 · → daily
value: qty=-35.6% · date=2026-06-15
[图: Qwen模型在Rallies Arena上的投资组合业绩表现与标普500走势对比图 — Portfolio Value: $64,395; Total Return: -35.60%; SPY Return: +9.91%; Market Beat: -45.51%; Max Drawdown: -37.7%]
📷 原图
<a id="atom-e8c1eed6e1844754"></a>🟦 2026-06-15 X·@wallstengine · 投资组合价值 · 仓位情绪
Portfolio Value: $64,395
tags: 好数字 · → daily
value: qty=$64,395 · date=2026-06-15
[图: Qwen模型在Rallies Arena上的投资组合业绩表现与标普500走势对比图 — Portfolio Value: $64,395; Total Return: -35.60%; SPY Return: +9.91%; Market Beat: -45.51%; Max Drawdown: -37.7%]
📷 原图

⏱ 时间轴 (近 20)

🟥 2026-06-29 · forecast · 预期在接下来几个月会突破模型能力阈值 · → daily
🟥 2026-06-28 · fact · 本地模型vs闭源 · → daily
🟦 2026-06-27 · fact · 最常见行为是拒答/回避 · → daily
🟦 2026-06-27 · fact · 定价低于美国竞品 · → daily
🟥 2026-06-26 · narrative · 200-300B参数模型效果优秀 · → daily
🟦 2026-06-26 · fact · 模型发布 · → daily
🟥 2026-06-26 · narrative · run through an abliteration now · → daily
🟥 2026-06-25 · narrative · has the vast majority of market share · → daily
🟦 2026-06-25 · fact · 招聘多模态智能体实习生 · → daily
🟥 2026-06-25 · narrative · model release frequency · → daily
🟦 2026-06-24 · fact · 发布了 AgentWorld · → daily
🟦 2026-06-24 · fact · released native language world models that simulate 7 agenti · → daily
🟦 2026-06-24 · fact · trained on 10M+ trajectories · → daily
🟦 2026-06-24 · fact · releasing 35B-A3B and 397B-A17B MoE weights plus AgentWorldB · → daily
🟥 2026-06-23 · fact · 写作capability · → daily
🟦 2026-06-23 · fact · open-weight release plan · → daily
🟦 2026-06-23 · fact · 发布世界模型基准 · → daily
🟦 2026-06-23 · fact · 基准构建于前沿模型轨迹 · → daily
🟦 2026-06-23 · fact · 基准测试AI预测环境状态能力 · → daily
🟦 2026-06-23 · fact · 数据集包含2170轮评估 · → daily

← 实体目录 · 系统日志