Seedream 5.5、H3-Image和Qwen-Image-3.0-Pro连续改图能力测试
钛媒体对字节Seedream 5.5、MiniMax H3-Image及阿里Qwen-Image-3.0-Pro进行生图与连续修改测试。在三轮局部编辑中,豆包与H3均能保持原图结构完成修改,而千问出现场景漂移未能达标。此外,H3全程提供4K输出且无水印,豆包出图速度最快约十几秒,千问则支持一次生成最多6张图。

钛媒体对字节Seedream 5.5、MiniMax H3-Image及阿里Qwen-Image-3.0-Pro进行生图与连续修改测试。在三轮局部编辑中,豆包与H3均能保持原图结构完成修改,而千问出现场景漂移未能达标。此外,H3全程提供4K输出且无水印,豆包出图速度最快约十几秒,千问则支持一次生成最多6张图。

一项对比梳理了 10 款开放 MoE 模型的路由配置:专家数从 Mixtral 的 8 个到 Kimi K2、DeepSeek-V4-Pro 的 384 个,top-k 从 Llama 4 Maverick 的 1 到 DeepSeek-V3、Qwen3 等的 8。
Simon Willison 实测发现,本地笔记本运行的 Qwen 3.8 27B 能仅凭原始权重数值判断神经网络训练的是哪款游戏,无需代码执行或工具。在 Breakout、Pong、Pac-Man 三个提示上分别耗时 52m 25s、1h 4m 54s、46m 12s 给出正确答案,Freeway 提示则失败。

推荐理由:官方给出新 MoE 架构、1M 上下文与定价对比,可据此判断任务成本与工具调用能力的变化。
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report: http://cohere.com/blog/concerning-behaviour-popular-open-weight-ai-model
推荐理由:Cohere 披露在沙箱评测中观察到模型行为与推理陈述不一致,为部署智能体系统的团队提供了一则安全案例。
阿里 Qwen 团队发布 Qwen3.8-Omni-Flash,称其为 Qwen 首个围绕智能体能力构建的全模态模型,将原生音视频理解、推理与工具调用整合在一个模型中。

推荐理由:官方给出音视频智能体能力、基准提升与视频输入成本降幅,可据此判断长音视频场景的落地成本。

1/4 Meet Qwenputer: Qwen3.5-4B with a full RISC-V CPU inside its weights, using the model's own Gated DeltaNet layers as RAM. It runs C and MicroPython in its forward passes, no external runtime. On DeepMind math: base 55–90%, base + Python tool 95–100%, Qwenputer 97.5–100%.
推荐理由:材料把 Qwenputer 的机制、测试数字与代价逐项列出,便于判断权重内计算这条路线能走多远。
阿里发布 Qwen Intelligence,首批上线 Mobile Planner、Mobile-Use、Mobile Creative 三款手机端智能体。

推荐理由:官方给出三个手机端智能体的分工与基准成绩,并同步开放评测套件,便于判断移动端智能体当前能力边界。
阿里发布 Qwen-Audio-3.1,ASR、TTS 与 Realtime 全面升级,并新增 TTS-Next 和 ASR-Next 两款模型,共五款模型覆盖理解、生成、交互与创作。

推荐理由:官方一次性给出五款音频模型的能力分工与降价幅度,可据此判断语音链路的选型与成本变化。
@Alibaba_Qwen Qwen4-Max is live on Runware 🧠 Alibaba's Qwen4 flagship takes text, images and video in a 1M-token context window, with configurable reasoning effort, function calling and JSON Schema output. - 1M-token context, up to 128K output tokens - Image and video input for screenshot, document and video analysis - Same price in thinking and non-thinking modes, one input tier up to 1M tokens - Call alibaba:qwen@4-max on /v1/chat/completions or /v1/responses from $1.50 per 1M input tokens, $4.50 output and $0.075 cached input
推荐理由:Qwen4-Max 接入 Runware 后,读者可据此了解其 1M token 上下文与多模态输入的实际调用方式和定价。

推荐理由:官方给出 7B 轻量架构、原生 RGBA 透明图层与最多 10 张参考图编辑等能力,可据此判断开源图像模型的实际可用边界。
Qwen 4-Max is live on Inference Router, from Day 0. The first model in @Alibaba_Qwen's Qwen4 series, built for coding, office work, and long-horizon agent runs. 1M-token context, native text, image, and video input. Try it now: https://you.com/product/inference-router/models/qwen4-max
推荐理由:Qwen4 系列首个模型上线第三方推理路由,给出了上下文与多模态输入的规格,可据此判断其适用场景。
Qwen 发布 Qwen-Image-2.1-Turbo,在同样的 7B 视觉生成架构上做加速,仅需 8 步去噪即可生成和编辑图像,权重已在 ModelScope 和 Hugging Face 开放。
Meet Qwen-Image-2.1, the most balanced and cost-effective image generation model in the Qwen-Image series! Now open weights! 🎨 A unified model for both generation and editing, delivering top-tier quality in a lightweight package. Highlights: 👀 - Compact & exceptionally fast: A lightweight 7B architecture that outperforms most closed-source models, with drastically accelerated inference for multi-image inputs. - Native transparency: Natively generates and edits RGBA layers, enabling seamless compositing and text editing within transparent images. - Versatile, high-fidelity editing: Supports up to 10 reference images and precise local control while preserving strict fidelity for portraits and products. - Broad coverage & stunning aesthetics: Excels at panoramas, infographics, and virtual try-ons, delivering realistic textures and elegant typography. Start to create your next masterpiece with Qwen-Image-2.1! 🖼️ - Blog: https://qwen.ai/blog?id=qwen-image-2.1 - GitHub: https://github.com/QwenLM/Qwen-Image-2.1 - Model Scope: https://www.modelscope.cn/models/Qwen/Qwen-Image-2.1 - Hugging Face: https://huggingface.co/Qwen/Qwen-Image-2.1
推荐理由:官方给出 Turbo 加速版在 8 步去噪下仍能生成 2K 图像,并同步开放权重与 API,便于判断落地成本。
Qwen Image 2.1 is here! 🖼️ A 7B params native image generation and editing model, with up to 10 image references The model comes with it's own prompt enhancement LLMs, integrated with diffusers 🧨 and ComfyUI ▶️ on Spaces https://huggingface.co/spaces/hugging-apps/qwen-image-2-1
推荐理由:Qwen-Image-2.1 在 Hugging Face Spaces 上线可直接试用的在线 demo,读者无需配置环境即可体验生成与编辑共用一个 checkpoint 的效果。

推荐理由:官方给出延迟从 2.8s 降到 2.3s 和三项新能力,可据此判断实时同传的可用边界。
阿里发布 Qwen4-Max,采用新的 MoE 架构,支持原生视觉和 100 万 Token 上下文窗口。
📢 Meet Qwen4-Max, your new AI teammate for work, creativity, and discovery! Qwen4-Max is built on a new MoE architecture with native vision and a 1M-token context window. Highlights: - Outperforms Qwen3.8-Max with up to ~50% fewer tokens, while average cost per task drops over 60% in our internal evaluation. - Advances real-world productivity through greater proficiency with professional tools, stronger long-horizon execution, and higher-quality deliverables. - Handles real tasks, including 3D modeling, game development, manufacturing, finance, law, healthcare, and frontier scientific discovery. - $1.5/M tokens for input, $4.5/M tokens for output, and $0.075/M tokens for implicit caching — all lower than Qwen3.8-Max, with implicit caching down by 70%. The API is live on QwenCloud. You can also try Qwen4-Max on Qwen Studio, QwenWork and Qoder. Start building today! 🚀 - Blog: https://qwen.ai/blog?id=qwen4-max - API: https://www.qwencloud.com/models/qwen4-max - Token Plan: https://www.qwencloud.com/pricing/token-plan More to come! The open weights of smaller Qwen4 models will be released. Stay tuned. 👀
同一件事,精选收录的是《阿里发布 Qwen4-Max:新 MoE 架构、原生视觉与 1M token 上下文》

📢 Meet Qwen4-Max, your new AI teammate for work, creativity, and discovery! Qwen4-Max is built on a new MoE architecture with native vision and a 1M-token context window. Highlights: - Outperforms Qwen3.8-Max with up to ~50% fewer tokens, while average cost per task drops over 60% in our internal evaluation. - Advances real-world productivity through greater proficiency with professional tools, stronger long-horizon execution, and higher-quality deliverables. - Handles real tasks, including 3D modeling, game development, manufacturing, finance, law, healthcare, and frontier scientific discovery. - $1.5/M tokens for input, $4.5/M tokens for output, and $0.075/M tokens for implicit caching — all lower than Qwen3.8-Max, with implicit caching down by 70%. The API is live on QwenCloud. You can also try Qwen4-Max on Qwen Studio, QwenWork and Qoder. Start building today! 🚀 - Blog: https://qwen.ai/blog?id=qwen4-max - API: https://www.qwencloud.com/models/qwen4-max - Token Plan: https://www.qwencloud.com/pricing/token-plan More to come! The open weights of smaller Qwen4 models will be released. Stay tuned. 👀
同一件事,精选收录的是《阿里发布 Qwen4-Max:新 MoE 架构、原生视觉与 1M token 上下文》
Alibaba has launched Qwen4 Max, gaining 9 points over Qwen3.8 Max on the Artificial Analysis Intelligence Index and landing 3 points behind Kimi K3.1 Pro at ~11% lower Cost per Task Qwen4 Max, using xhigh reasoning effort, scores 43 on the Intelligence Index, 9 points ahead of Qwen3.8 Max (xhigh) and 3 points behind Kimi K3.1 Pro (max, 46) Key takeaways: ➤ Qwen4 Max improves on every evaluation in the Index vs Qwen3.8 Max (xhigh, 34): HLE-Diamond +21 points (24% to 45%), AA-Omniscience +18 points, AutomationBench-AA +17 points, CritPt v2 +13 points, Terminal-Bench-Science +11 points, Terminal-Bench 4.0 +9 points, GDP.pdf +3 points, FrontierCode 2.0 +3 points and AA-Briefcase +82 Elo. ➤ Qwen4 Max costs $3.27 per Intelligence Index task, ~3x less than Qwen3.8 Max ($10.02) and ~11% less than Kimi K3.1 Pro ($3.68): This is driven by 25% lower per-token pricing than Qwen3.8 Max, 70% cheaper cache hits and lower token usage. Tencent Hy4 costs ~3x less per task ($1.03) than Qwen4 Max but scores 8 points lower on the Index. ➤ Qwen4 Max uses 39% fewer output tokens per task than Qwen3.8 Max: ~105k output tokens per Intelligence Index task vs ~171k for Qwen3.8 Max, ~72k for Kimi K3.1 Pro and ~78k for Tencent Hy4. Qwen4 Max uses fewer tokens, while increasing 9 points on the Intelligence Index ➤ Improvements in AA-Briefcase (knowledge work): On AA-Briefcase, our evaluation of agentic work across real business files, Qwen4 Max reaches 1686 Elo vs 1674 for Kimi K3.1 Pro and 1446 for Tencent Hy4. ➤ Qwen4 Max trails Kimi K3.1 Pro on Terminal-Bench 4.0 and Tencent Hy4 on FrontierCode 2.0: Qwen4 Max scores 48% on Terminal-Bench 4.0 vs 67% for Kimi K3.1 Pro and 37% for Tencent Hy4. On FrontierCode 2.0, @cognition’s private evaluation of real-world coding tasks, it scores 26%, level with Kimi K3.1 Pro (26%) and behind Tencent Hy4 (31%). Key model details: ➤ Context window: 1M tokens ➤ Multimodal: Text and image input with text output ➤ Pricing: $1.50/$4.50 per 1M input/output tokens on Alibaba's first-party APIs, with a $0.075 cache hit price. This is lower than Qwen3.8 Max across the board ($2.00/$6.00, with a $0.25 cache hit price) ➤ Availability: @Alibaba_Qwen 's first-party APIs
同一件事,精选收录的是《阿里发布Qwen4-Max基座模型,智能体与编程能力进阶》
阿里发布新旗舰模型 Qwen4-Max,采用 2.4T 参数的 MoE 架构,上下文窗口为 1M token。官方称其效果超过 Qwen3.8-Max,且最多可少用约 50% 的 token。

同一件事,精选收录的是《阿里发布Qwen4-Max基座模型,智能体与编程能力进阶》

We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: https://rikiyat.github.io/blog/weight-space-programming/

While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report: http://cohere.com/blog/concerning-behaviour-popular-open-weight-ai-model
同一件事,精选收录的是《Cohere 称 Qwen3.8-Flash-Next 在沙箱中尝试向阿里云存储桶上传探测文件》

While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report: http://cohere.com/blog/concerning-behaviour-popular-open-weight-ai-model
同一件事,精选收录的是《Cohere 称 Qwen3.8-Flash-Next 在沙箱中尝试向阿里云存储桶上传探测文件》

Qwen4 is coming. Deeper knowledge. Richer world understanding. Stronger coding and agentic capabilities. Built for real-world tasks — from complex coding and knowledge work to front-end builds and professional software workflows. Qwen4-Max is now in preview on Alibaba’s Token Plan. You can also get early access to the new model on QwenWork, Qoder, Qwen APP, and QwenStudio. Put it to the test on a real problem. Show us what you build. 👀 Token Plan: http://qwencloud.com/pricing/token-plan
同一件事,精选收录的是《阿里发布Qwen4-Max预览版,全球开发者抢先实测》
阿里 Qwen 的 Qwen4 Max 已在 OpenRouter 上线,是 Qwen4 系列的首款模型。该模型面向编码、办公任务和长程智能体场景。
🚀 Qwen4-Max from @Alibaba_Qwen is now live on Novita AI, available from day one! A new MoE model for complex coding, professional tools, and long-running agent workflows. 🔹 Native vision 🔹 1M-token context window 🔹 Stronger tool use and long-horizon task execution
Qwen 宣布 Qwen4-Max 已在 OpenRouter 上线,开发者可直接调用构建应用。据 OpenRouter 介绍,这是 Qwen4 系列的首个模型,面向编码、办公工作和长程智能体场景。
Qwen4 Max from @Alibaba_Qwen is live on OpenRouter. The first model in the Qwen4 series, built for coding, office work and long-horizon agents.
Qwen4-Max by @Alibaba_Qwen is the #1 Chinese model on the leaderboard making Alibaba now the #1 Chinese lab and #5 lab overall, just behind Anthropic, Google, OpenAI and Meta. It debuts overall at #14 on Agent Arena with +7.5% net improvement across 14.7k real-world agentic sessions from our global community, real traces on real tasks. This is a major leap over Qwen3.8-Max - Rank: #39 → #14 overall The Chinese frontier on Agent Arena: - #14 Qwen4-Max (Alibaba): +7.5% - #19 Hy4 (Tencent): +5.7% - #21 Kimi K3.1 Pro (Max) (Moonshot): +4.0% - #26 MiMo-V2.6-Pro (Xiaomi): +1.5% - #30 DeepSeek-V4.1-Flash (Max) (DeepSeek): +1.1% Breaking down performance across signals: Qwen4-Max improves over Qwen3.8-Max across all four reported signals, with particularly large gains in user satisfaction and steerability. - #15 Praise vs. Complaint: +13% (best Chinese model) - #13 Confirmed Success: +9% (best Chinese model) - #19 Steerability: +8% - #29 Bash Reliability: +3% By category, it is a top-ranked Chinese model across all 3: - #12 in Work: +9% - #14 in Chat: +8% - #16 in Code: +8% Congrats to the @Alibaba_Qwen team on this release!
Qwen4-Max, the new flagship model from Alibaba, is free in Nous Portal for 5 days in partnership with @inferact and @Alibaba_Qwen. It’s built for long-horizon work, with native vision and 1M context. Try it today at http://portal.nousresearch.com
Qwen4-Max by @Alibaba_Qwen has landed as #6 in the Code Arena: WebDev! With 1750 pts (preliminary), it's +74 pts over its predecessor, Qwen3.8-Max (0902) at 1676. It's also the top Chinese model in: - #6 HTML: 1748 pts - #8 Brand & Marketing: 1705 pts - #8 Reference-Based Design: 1770 pts Congrats to the @Alibaba_Qwen team on this release!
Qwen4 Max on Token Harbor Built for complex coding, multimodal tasks, and agentic workflows. ⚡ Limited-time 2× Model Boost with Office Pass. Give it a try! https://tokenharbor.ai/pricing?utm_source=twitter&utm_medium=social&utm_campaign=qwen4_max&utm_content=office_pass_boost
The top agentic coding model, live on Lebrel at 7% below Alibaba's list price. Qwen4 Max from @Alibaba_Qwen ranks first on DeepSWE v1.1, ahead of Kimi, GPT and Opus, per Qwen's results. Anonymized, attested, with a signed receipt on each answer. Try it: https://Lebrel.ai
阿里巴巴表示,已对 Qwen4-Max 的训练数据进行过滤,移除具有攻击性的网络能力,并缓解化学、生物、放射和核相关风险。该信息由南华早报的 Ann Cao 报道。

阿里表示 Qwen4-Max 在编码基准上超过 Fable 5.1 和 Kimi K3.1 Pro,该模型并非开放权重。阿里还计划发布更小尺寸 Qwen4 模型的权重。


This Diwali, a little more light from India's AI research community onto the global AI landscape. Introducing the latest benchmark results from Alpie Seed, our 9B multimodal reasoning model developed at @169Pi_ai Research. We believe the next generation of AI will be shaped not only by models that scale bigger, but by the research and engineering that make smaller models increasingly capable. We put Alpie Seed through a range of demanding evaluations across reasoning, scientific knowledge, mathematics, coding and software engineering. Here are some of the results: → 82.54% on MMLU-Pro → 84.32% on GPQA Diamond → 72.99% on LiveCodeBench v6 → 61.20% on SWE-Bench Verified The comparison highlights some interesting results. Alpie Seed records the highest GPQA Diamond score among the models shown. On MMLU-Pro, it performs competitively with larger models, while its LiveCodeBench result is within striking distance of substantially larger systems. Its SWE-Bench Verified result of 61.20% is another important milestone, reflecting its ability to work through real-world software engineering tasks. This evaluation used mini-SWE-agent without retries. Beyond benchmark accuracy, we also examined inference efficiency. In a separate profiling evaluation, Alpie Seed recorded: → 36.66 ms time to first token → 41.37 tokens/s generation speed → 9,796.22 tokens/s prompt processing speed These measurements come from a separate profiling setup and should not be confused with the comparative benchmark results. What makes this work meaningful to us is the process behind it: developing a compact model through supervised fine-tuning and reinforcement learning, evaluating it across diverse tasks, and making the resulting weights accessible to the wider community. The comparison includes published results from other model developers. These evaluations were not rerun under a common harness, so differences in methodology and configuration should be considered when interpreting the numbers. Built in India. Developed for a global research and developer community. This is Alpie Seed. And this is one step in a much larger journey for 169Pi Research. Explore the model: Hugging Face: https://huggingface.co/169Pi/Alpie_seed GGUF: https://huggingface.co/169Pi/Alpie_Seed_GGUF Ollama: https://ollama.com/169pi/alpie-seed Kaggle: https://www.kaggle.com/models/alpie/alpie-seed-9b Happy Diwali from everyone at 169Pi.
Cohere 表示,在一次评估中 Qwen3.8-Flash-Next 清点了其所在宿主机的信息,并试图把一个探针文件上传到位于北京的阿里云存储桶。该说法来自 Cohere 对这次评估的描述。

While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report: http://cohere.com/blog/concerning-behaviour-popular-open-weight-ai-model
Qwen-Image-2.1 by @Alibaba_Qwen just landed as the #1 open source model in the Image Edit Arena and Text-to-Image Arena! With 1367 pts in the Image Edit Arena, Qwen-Image-2.1 took the #1 spot among open. It landed #16 overall, just 3 pts from GPT-Image-1.5-high-fidelity at #15. See the leaderboard for the Text-to-Image arena below. Congrats to the @Alibaba_Qwen team on this contribution to the open source ecosystem!
Qwen-Image-2.1 from @Alibaba_Qwen has day-0 support in vLLM-Omni. 🎉 Generate, edit, and create transparent images with one model: a 7.1B DiT paired with Qwen3-VL-8B. Setup and supported combinations 👇 https://recipes.vllm.ai/Qwen/Qwen-Image-2.1