@Alibaba_Qwen @OpenRouter When can I use it in coding plan ?
English
Charlie Chen
20 posts
Qwen3.6-Plus ranks # 1 on @OpenRouter , and the first model on OpenRouter to break 1 Trillion tokens processed in a single day！!🥇🔥
We are thrilled to see Qwen3.6-Plus topping the charts so quickly. This milestone wouldn't be possible without our amazing developers. ❤️Thank you！！
OpenRouter@OpenRouter
Qwen 3.6 Plus from @Alibaba_Qwen is officially the first model on OpenRouter to break 1 Trillion tokens processed in a single day! At ~1,400,000,000,000 tokens, it’s the strongest full day performance of any new model dropped this year. Congrats to the Qwen team!
English
@alibaba_cloud @OpenRouter when can I use it in qwen coding plan ?
English
Qwen3.6-Plus ranks # 1 on @OpenRouter, and the first model on OpenRouter to break 1 Trillion tokens processed in a single day! 🥇🙌 At ~1,400,000,000,000 tokens, it's the strongest full day performance of any new model dropped this year. We are thrilled to see Qwen3.6-Plus topping the charts so quickly. This milestone wouldn't be possible without our amazing developers. ❤️ Thank you! !#Qwen #OpenRouter #AI #NumberOne #LLM #AInnovation
English
@apanasenko @sydneyrunkle Could you please elaborate on what you mean by 'violates the rules of constraint sampling for the GPT family'?
English
@sydneyrunkle It’s a very poor practice that breaks prompt caching and violates the rules of constraint sampling for the GPT family.
English
day 2 of the harness engineering series: dynamic config
middleware lets you reshape your agent's model, tools, and prompt at every step based on context.
ex: LLMToolSelectorMiddleware runs a fast filter on your tool registry so your main model receives streamlined tool specs.
English
OpenClaw的企业级管理面板工具：ClawManager，一人装全员用，等于公司IT部门的管理后台
想在团队/公司内部规模化部署OpenClaw的可以看看
ClawManager是一个面向团队的Kubernetes控制面板，用于统一管理OpenClaw和Linux桌面运行时
一个管理后台统一管理用户、配额、实例和运行时镜像
资源配额可控，比如每个人用多少CPU、内存、GPU等
所有桌面跑在集群内部，同事通过平台认证后才能访问，不直接暴露Pod端口
AI网关治理，为OpenClaw实例提供统一的OpenAI兼容入口，叠加模型管理、审计追踪、成本核算和风控策略（可自动拦截或路由到安全模型）
集群资源总览，节点、CPU、内存、存储等资源可视化
OpenClaw记忆/偏好备份支持导入导出
#openclaw #ClawManager
中文
@Zai_org @NVIDIAGTC 现在已经在日常的开发工作中使用 GLM 系列模型+Speckit来协助完成从0-1的新需求，单元测试 及 code review 等开发活动。用 GLM-5 完成的单元测试比其他开源模型一次性完成的能力更高。唯一的困扰就是 token 使用过快，经常受到限制。😄
中文
🤗To celebrate @NVIDIAGTC and the launch of GLM-5-Turbo, we're running a giveaway!
Our team will be picking lucky winners on a rolling basis. You have a chance to get a free month of the Max Coding Plan!
Here's how to join in: a retweet, reply, or post with:
1. GLM-5-Turbo usecases or
2. A short write-up based on your experience or honest take
The window closes in 48 hours. All prizes will be sent out within 72 hours after that. Feel free to get your User ID here: z.ai/subscribe
Please jump in, we'd love to see what you're building!
ps. @louszbd is at GTC right now.😆 Come say hi in person!
English
@llama_index As a developer using the LlamaIndex agent framework, I’m very interested in how to implement something similar with it. Are there any tutorials available?😉
English
🚀 LlamaAgents Builder just leveled up: File uploads are here!
Our natural language interface for building agentic document workflows now supports file uploads.
You can provide example documents as context, and the agent will use them as a starting point to design and tailor your workflow. The result? Applications that better match your real-world use case.
The more representative your sample files, the more accurate your final app.
🎥 Watch the full walkthrough: youtu.be/5Nk6KZhBDbQ
🦙 Get started with LlamaCloud: cloud.llamaindex.ai/signup
YouTube
English
Introducing Pika AI Selves: AI you birth, raise, and set loose to be a living extension of you. They’re rich, multi-faceted beings with persistent memory, and maybe even a peanut allergy. It’s up to you!
Have them send pictures to your group chat. Make a video game about your fish. Call your mom while you do anything but call your mom. The possibilities are as myriad as the stars ✨
Get on the list to give birth to yours at pika dot me
English
@cline Looks interesting, but still don't get why terminal ai is the future？why not agentic IDE?
English
Introducing Cline CLI 2.0: An open-source AI coding agent that runs entirely in your terminal.
Parallel agents, headless CI/CD pipelines, ACP support for any editor, and a completely redesigned developer experience. Minimax M2.5 and Kimi K2.5 are free to use for a limited time.
From prompt to production. All in your terminal.
English
@Hesamation @huggingface @Thom_Wolf @_lewtun @LoubnaBenAllal1 @Nouamanetazi @eliebakouch @edwardbeeching @lvwerra The link is dead. Failed to retrieve error logs: SSE is not enabled
English
holy shit... Hugging Face cooked again! 🔥
they just dropped a free blog (BOOK) that covers the no-bs reality of building SOTA models. i haven't seen any lab/researcher go into the real decisions behind the LLM research and its nuances. this is literally a gem.
Syllabus:
→ Training compass: why → what → how
→ Every big model starts with a small ablation
→ Designing the model architecture
→ The art of data curation
→ The training marathon
→ Beyond base models — post-training in 2025
→ Infrastructure - the unsung hero
skimming through the blog, this is incredibly detailed just like their ultrascale playbook. i'm gonna read this and share more about it in the coming days.
Read here: huggingface.co/spaces/Hugging…
English
@tarat_211 @openrouter Why is the latency of GPT-5 so slow compared to other models with the reasoning_effort setting at minimal?
English
@akshay_pachaar Great list, however, for the 6th skill (i.e. Thinking tools-first, not model-first), could you explain more about it ?
English
15 underrated skills for AI engineers (that compound over time):
1. Writing effective prompts
2. Knowing when fine-tuning vs. not
3. Building reliable ML pipelines
4. Balancing speed vs cost vs accuracy
5. Reading research with a builder’s mindset
6. Thinking tools-first, not model-first
7. Estimating ROI on model upgrades
8. Designing human-AI interfaces
9. Evaluating probabilistic systems
10. Monitoring models in production
11. Planning for model failures
12. Containerizing AI workloads
13. Managing model versions and rollbacks
14. Optimizing inference costs effectively
15. Building ethical AI guardrails
English
强，阿里通义刚刚又放出了一款深度研究智能体：通义DeepResearch，30B参数媲美OpenAI Deep Research
Humanity's Last Exam得分32.9，BrowseComp得分45.3，xbench-DeepSearch得分75.0
128K上下文长度，擅长长周期、深度信息搜集，需要进行复杂问题分解、多步推理、信息搜集整合的场景可以用
其用智能体数据持续预训练的方式，来增强推理和规划能力；用on-policy强化学习方法，来确保决策能力的稳健性
两种推理模式：
ReAct 模式，单模型逐步推理，轻量
Heavy 模式，多Agent并行IterResearch，再统一合成答案，测试时算力可扩展
目前这个智能体已经落地到高德地图的多日行程规划，以及法律助手通义法睿里了
#深度研究 #DeepResearch #tongyiDeepResearch
中文
@alexatallah @simonw What's the differences between reasoning and reasoning_details in the response?
English
@simonw Note that OpenRouter preserves chain-of-thought under the hood when using chat completions
English
TIL that the OpenAI Responses API gives better performance for retaining models over Chat Completions because it better preserves their chain-of-thought throughout the ongoing conversation
prashant@prashantmital
myth #3: model intelligence is the same regardless of whether you use completions or responses wrong again. responses was built for thinking models that call tools within their chain-of-thought (CoT). responses allows persisting the CoT between model invocations when calling tools agentically -- the result is a more intelligent model, and much higher cache utilization; we saw cache rates jump from 40-80% on some workloads. this one is perhaps the most egregious. developers don't realize how much performance they are leaving on the table. i get it, its hard because you use LiteLLM or some custom harness you built around chat completions or whatever, but prioritizing the switch is crucial if you want GPT-5 to be maximally performant in your agents. here's our cookbook on function calling with responses: cookbook.openai.com/examples/o-ser…
English
Vercel AI Gateway is now generally available.
• Access hundreds of models
• Zero markup on tokens (including BYOK)
• No provider accounts needed
• High rate limits
• Failover for high reliability
• Sub-20ms latency
• AI SDK and OpenAI-compatible
vercel.fyi/ai-gateway
English
We're announcing 2 huge @firecrawl updates in 24 hours 🔥
You're not going to want to miss this... we're super pumped!
Stay tuned and comment "🔥" for some free merch after launch 👀
English
🚀 Meet Qwen-Image — a 20B MMDiT model for next-gen text-to-image generation. Especially strong at creating stunning graphic posters with native text. Now open-source.
🔍 Key Highlights:
🔹 SOTA text rendering — rivals GPT-4o in English, best-in-class for Chinese
🔹 In-pixel text generation — no overlays, fully integrated
🔹 Bilingual support, diverse fonts, complex layouts
🎨 Also excels at general image generation — from photorealistic to anime, impressionist to minimalist. A true creative powerhouse.
Blog:qwenlm.github.io/blog/qwen-imag…
Hugging Face：huggingface.co/Qwen/Qwen-Image
ModelScope：modelscope.cn/models/Qwen/Qw…
Github：github.com/QwenLM/Qwen-Im…
Technical report：…anwen-res.oss-cn-beijing.aliyuncs.com/Qwen-Image/Qwe…
Demo: modelscope.cn/aigc/imageGene…
English
@llama_index @tuanacelik it's amazing, but when will nosql be supported. It's not easy to migrate old memory block based on mongo to new one.🤔
English
There can be a number of different approaches to agent memory, each serving a different purpose. That's why we recently started to introduce flexible Memory Blocks to LlamaIndex:
- Fact extraction
- Static
- Vector memory
and more..
Next week, @tuanacelik will be on a livestream with @AIMakerspace to discuss everything agent memory, and our approach to it at LlamaIndex. Sign up to learn more:
lu.ma/t27lryii
English
@jerryjliu0 How about AGNTCY? What's the difference between them
English
I spent 10 mins trying to figure out what the "practical" difference between Google's A2A and MCP are.
What exactly is the difference between agent<>agent communication vs. agent<>tool communication?
* The only thing I arrived at was that agent<>agent communication allows two agents to collaborate with each other and send messages back and forth.
(there's nothing stopping a "tool" from doing this too with MCP's stateful protocol, but to properly process these messages implies an LLM in the loop which would make it "agentic" anyways)
Everything else - capability discovery, task management, seem like things MCP can also handle...eventually, even if not now
Thoughts?
Source: developers.googleblog.com/en/a2a-a-new-e…
English
牛，Mistral刚刚发布了号称地表最强OCR，给文档理解设立了新标准！
Mistral OCR具备强大认知能力，能准确理解文档中包括文本、图像、表格、公式等在内的每个元素
特点：
1、原生多语言和多模态，支持数千种文字、字体以及语言
2、能准确理解复杂的文档元素，包括图像、数学公式、表格以及 LaTeX 格式等，尤其擅长处理包含图表、图形、公式和插图的科学论文等富文档
3、在多个文档分析方面的基准测试中优于其他OCR模型，尤其在扫描文档、表格和数学公式识别上表现出色
4、处理速度很快，单节点每分钟可处理2000页
5、支持使用文档作为提示，以结构化格式比如 JSON输出
6、可选择性自托管
#OCR #MistralOCR #Mistral
中文