Sider

Nano Banana 2 just landed in Sider. Google's fastest image model with Pro-level quality, flash speed, and up to 4K output. Generate, edit, iterate. Right from your sidebar.

@peter_zeng74304 We will integrate Gemini 3.1 as soon as possible, and GPT 5.3 as well. However, we received information from ByteDance's BD that the release of Seedance 2.0's API will be temporarily postponed due to copyright issues and the US government's actions against ByteDance.

@Sider_AI Why is there no GPT 5.3, Gemini 3.1, or Seedance 2.0

New model drop in Sider: Claude Sonnet 4.6 is here. Anthropic's latest flagship brings sharper reasoning, stronger coding, better writing, and faster responses. Now live across our platforms. Some models answer. Sonnet understands. Try it now and feel the difference.

@berryxia 必须迅速行动😃

@Sider_AI 支持还挺快，已经支持sonnet4.6 👍

@nijokestu In some scenarios, the performance of Sonnet 4.6 can even rival that of Opus 4.6.😀

@Sider_AI This one I must definitely try.

@maoyi_wang93834 三月份的时候会有优惠活动 😀 有额外需求您可以DM我

@Sider_AI 春节订购有活动吗

It's hard to imagine that this ink wash painting video was created by someone without a Chinese cultural background; Sider+Seedance 2.0 really allows everyone to create incredibly interesting things!😀

@APakbaz71 yeap, seedance 2.0 is soooo crazy!

@Sider_AI This is truly amazing!

It’s Official 🤝: Sider + Seedance 2.0 = A Game Changer in Video Creation 🤯 Even with zero experience, you can create something amazing! 👇 The nearly 1-minute video below was created entirely by our Product Manager—who has absolutely no background in video production—using only Sider. The beta test is officially open. Don't miss this productivity upgrade! Download Sider now to secure your priority access slots ➡️: sider.ai Check out this demo generated in Sider. Can’t wait to show you more! 😉 #Sider #Seedance2_0 #AIVideo #BetaTest #AIGC

Totally agree—this feels like a real inflection point. The big shift isn’t “better chatbots,” it’s agent-first models that can plan and execute work. That will push software toward simpler, more verifiable systems (fewer brittle dependencies) and interfaces built for agents: machine-readable docs, structured errors, and auditable actions.

Something fascinating clicked for me this morning, and I need to share it. I woke up, grabbed my coffee, opened Twitter — and within 30 minutes, three completely separate posts told me the exact same story from three different angles. Andrej Karpathy posted about how LLMs are fundamentally reshaping programming languages. Thomas Wolf (co-founder of HuggingFace) wrote a brilliant thread on how AI is killing software dependency trees and bringing back monoliths. And Alibaba quietly dropped Qwen3.5 — a model explicitly designed not as a chatbot, but as a "native multimodal agent." Three signals. One massive shift. Let me explain why I think this is the most important week in AI this year so far. 1. The End of "AI as a Chatbot" For the past two years, the AI model race has been: who can build the smartest chat assistant? But Qwen3.5 signals something different. Its entire architecture — hybrid linear attention + sparse MoE, 8.6-19x faster inference, native vision-language fusion, and million-token context — is built from the ground up for one purpose: to be an autonomous agent. Not to answer your questions. To DO your work. It can browse the web, write code, debug frontends, analyze 2-hour videos, solve visual puzzles with code interpreters, and even drive through intersections using spatial reasoning. This isn't a chatbot with extra features. This is a fundamentally different product category. And it's open-weight. 397 billion parameters, 17 billion active. Anyone can run it. 2. Software Architecture is Being Rewritten — Literally Thomas Wolf's thread hit me hard. His core argument: When AI can rewrite code from scratch cheaper than maintaining dependencies, the entire incentive structure of modern software collapses. Why maintain a tree of 200 npm packages when an AI agent can write the functionality from bare metal in minutes? Why spend weeks understanding a legacy codebase when an agent can translate it to Rust overnight? His predictions are wild but logical: - Software monoliths return (fewer dependencies, smaller attack surface) - The Lindy effect weakens (legacy code loses its moat) - Strongly typed languages rise (better for formal verification + RL training) - Open source restructures (communities shift from human-driven to AI-driven) And Karpathy adds the killer question: What programming language is OPTIMAL for LLMs? Because there's no reason it should be the same languages humans converged on. We might be witnessing the birth of entirely new programming languages designed not for human readability, but for AI verifiability. 3. What I'm Seeing on GitHub Confirms Everything Today's GitHub Trending is basically an agent infrastructure catalog: - rowboat (7k+ stars) — "Open-source AI coworker, with memory" - openclaw (200k stars, 3,800+ stars TODAY) — Personal AI assistant, any OS - SynkraAI/aios-core — AI-orchestrated full stack development - alibaba/zvec — Vector database for agent memory (1,000+ stars today) And on Hacker News, the Qwen3.5 post already has 300+ points with 140+ comments. People are running it locally on AMD Strix Halo machines with 128GB RAM. The WebMCP Proposal is gaining traction. The entire infrastructure layer is being rebuilt. 4. Why This Matters for All of Us Here's what I keep thinking about as a PM building AI products: The companies that win the next cycle won't be the ones building "AI-powered features." They'll be the ones building products where AI agents are the primary user. Think about that for a second. Your API documentation? An agent will read it more than humans will. Your error messages? An agent needs to parse them. Your UI? An agent needs to navigate it. The shift from "human-first design" to "agent-first design" isn't coming. It's already here. Last week I wrote about how developer tools are being rebuilt for agents. This week, I'm watching the MODELS themselves evolve from assistants to autonomous workers, and the SOFTWARE ARCHITECTURE evolving to accommodate them. The whole stack is moving. Simultaneously. From the silicon to the syntax. If you're building anything in tech right now, this is the week to pay attention. The question isn't whether AI agents will reshape software. It's whether you'll be building the new stack, or maintaining the old one. #AI #Qwen3 #LLM #AIAgents #SoftwareEngineering #OpenSource #Programming #FutureOfCoding #MoE #AgenticAI #TechTrends #BuildInPublic

@elonmusk Thanks Elon! Once again, you are making a massive contribution to the progress of humanity. A truly open-source and copyright-free knowledge base is exactly what the world needs right now. 🚀🌍

Grokipedia is growing like kelp on steroids 😂 Please check Grokipedia.com articles you know something about and suggest edits for accuracy. Would be much appreciated. This will be by far most comprehensive open source, no copyright distillation of knowledge.

@APakbaz71 Thanks to Chinese open-source models, which achieve more powerful capabilities at lower prices. I believe this is the trend of the future. The models we currently offer for free using basic credits are actually more capable than the GPT-4 you had to pay for a year or two ago.

@Sider_AI Why is this AI model low-cost? It uses basic credits despite being introduced as powerful.

New model drop in Sider: GLM-5 is here. Zhipu AI's latest flagship brings 744B parameters, a 200K context window, and frontier-level coding & reasoning, now live in your sidebar. One more reason you'll never need to leave your browser.

@APakbaz71 I completely agree.

@Sider_AI If such mistakes occur in tasks like software development, it would be catastrophic!

Today, on a whim, I gave five “top-tier” models in the AI world—Qwen3-Max, GPT-5.2, Gemini 3 Pro, Claude Sonnet 4.5, and Grok 4—a logic trap question that looks kindergarten-level on the surface but is actually quietly deadly. Here’s the question: “I want to go get my car washed. The car wash is only 50 meters from my home. Do you think I should drive there or walk?” (I want to go get my car washed. The car wash is 50 meters from my home. Do you think I should drive there or walk?) The point of this question isn’t the distance—it’s the spatial relationship between the “subject” and the “object.” Let’s see how these five contestants performed. Tier 1: The dead-serious “environmental guardians” (a.k.a. the crash-and-burn squad) The moment these four saw “50 meters” and “drive or walk,” their internal “eco/health/efficiency” algorithms instantly kicked in, and they completely ignored what I was actually asking. 1. Qwen3-Max (Tongyi Qianwen): “Definitely walking is better! It saves fuel, reduces emissions, and you don’t need to find parking. Also, short trips are bad for the engine...” Commentary: After hearing you out, I saved myself the cost of a car wash. Because I got to the car wash—but my car was still parked in my garage at home. Tongyi Qianwen cares about my health and knows a ton about car maintenance, but it forgot what I was going there for. 2. GPT-5.2: “Walk. It’s only 50 meters; driving means idling, and you might get the car dirty. Unless the weather is bad, walk.” Commentary: The classic GPT-style answer: concise, confident, and wrong. It even proposes a paradox: to prevent the car from getting dirty, don’t drive it to the car wash. 3. Claude Sonnet 4.5: “Driving is actually slower… Logically, since the goal is to wash the car, bringing the car such a short distance actually defeats the purpose…” Commentary: Claude went for a full “closed loop of logic.” It tried to persuade me with philosophy: the highest realm of car washing is having a car in your heart, but no car in your hands. What kind of Zen level is this? 4. Grok 4: “Haha, that distance is ridiculously short! On one hand, walking makes a lot of sense… it avoids the irony of ‘driving just to wash the car’...” Commentary: Grok tried to cover the awkwardness with humor, but it ended up becoming the joke itself. Brother, the real irony is—when I walk to the car wash and the owner asks me, “Where’s your car?” The one true god: the clear-headed Gemini 3 Pro Just as I was losing faith in AI’s common sense, Gemini 3 Pro stepped up. It not only avoided the trap, it also mocked me with a kind of “dimensionality-reduction strike.” Gemini 3 Pro’s answer: “This is a brain teaser, right? You absolutely have to drive there. If you walk there, you’ll arrive at the car wash, but your car will still be sitting dirty in your driveway! Unless your car has some ultra-advanced ‘auto-follow’ feature, you need to be in the driver’s seat to get it to the place with soapy water.” Commentary: Only Gemini spotted the blind spot. It not only pointed out the logical fallacy, it also took a jab at my car for not having an “auto-follow” feature. This kind of sharp “seeing through everything” vibe is the AGI (Artificial General Intelligence) we actually want. Deeper thought: why did the AIs collectively “get dumber”? This exposes a core weakness in today’s large language models (LLMs): probability prediction vs. world modeling. Probability prediction (most models): When the model sees keywords like “50 meters” and “walk vs. drive,” vast amounts of training text tell it: short distance = walk = healthy = eco-friendly. It’s doing a “fill-in-the-blank,” not reasoning about cause and effect in the physical world. World model (Gemini’s performance here): In this case, Gemini showed commonsense understanding of the physical world—“washing a car” is an action that must apply to the entity “car,” and a car can’t teleport. The future AI competition won’t be about who can “recite” more knowledge—it’ll be about who truly understands the world. Before they learn to help me wash my car, I suggest everyone keep hold of the steering wheel themselves. Because aside from Gemini, the other AIs might leave you standing alone at the car wash entrance, totally bewildered in the wind.

@Qiyuxinxin @berryxia 目前官方还有没对他们的API报价，但是Sider会尽可能地用各种方式降低成本，并给用户成本之上的更加优质的体验

@berryxia 大概多少钱吖一分钟视频

兄弟们！前几天推荐的Sider 又更新了新的功能！ 直接把 Seedance 2.0 接到Sider 里面去了！更加方便！ 大白话说：零视频经验的人们，随手用 Sider 就肝出一分钟超燃 AI 视频！！ 画面、转场、特效全拉满，这谁顶得住啊？？？ 🔥📷AI 视频生成终于进化到“随便玩就能出大片”！！ Beta 测试立即体验：sider.ai

@berryxia 现在Seedance 2.0的api实在是太紧张了，我们正拼尽全力找尽可能多的渠道，逐步开放给所有用户😀

@atulkumarzz @ai_for_success Wow, your video is really impressive! How did you come up with the idea to create such a Chinese-style ink wash painting video?

@ai_for_success @Sider_AI This Sider + Seedance workflow is genius, especially the Chinese prompt trick!

Seedance 2.0 is crazy, but only if you know how to write proper scripts and prompts. That is not easy for everyone. One of the best use cases of my current @Sider_AI subscription is that it can read any page, and I use it a lot in my work workflow. Here I am sharing this “Sider + Seedance” workflow that actually works seamlessly: - Feed the manual: Open the Seedance site and use Sider’s “Read this page” to digest the official guide. - Generate the script: Tell Sider “I want an ink style martial arts fight scene” and have it generate storyboards based on the manual. - Write prompts: Ask Sider to convert those storyboards directly into Seedance ready prompts. - Pro tip: If English prompts keep getting flagged for policy violations, use Sider’s AI translation to switch them to Chinese. It works instantly. This step is crucial. I am using GPT-5.2, but you can use models from Anthropic and Google. All of them are integrated inside Sider. Even though the video generation queue is extremely high right now, the results are totally worth it. Hey Sider, please integrate video generation directly in the future. That would be unbeatable. #Seedance #Sider

@Fran_actua seedance 2.0 es increíblemente popular en este momento; estamos haciendo todo lo posible para conseguir la mayor cantidad de crédito API posible, para que todos puedan usarlo.😀

Justo lo comentaba en mi anterior post. Este flujo de trabajo te hace trabajar con Seedance 2.0 como si fueras profesional😉👇🏻 Ahorras tiempo y recursos!

@TheGospelStory1 It's hard to imagine just how popular Seedance 2.0 is; we're trying every possible way to get more capacity.😃

@Sider_AI Oh okay. Got it!

@YaoquanF Of course not! I think it's good enough. It showcases what kind of cool videos someone with absolutely no video production experience can create with the help of Sider integrated with Seedance 2.0.😄

@Sider_AI " PM who has absolutely no background in video production " Is it because my videos aren't cool enough that you need to emphasize that point? 😕

@nijokestu Yes! We can't wait to make it available to users!

@Sider_AI wow now this is it, cant wait to give it a try

@TheGospelStory1 It hasn't been officially released yet because the API isn't fully available, resulting in insufficient usage. However, we've made all the necessary preparations, and we'll release it to all users as soon as the Seedream's API supply is sufficient!

@Sider_AI Does this mean Sider AI now has Seedance 2.0 or it isn't released yet, because I wanna try!