This is big for all, GPT image 2 is added now on Commonstack Model numbers are kept increasing on @commonstack_ai Remember we are still building, believe on @Gradient_HQ

GPT Image 2 is now officially available on @commonstack_ai This is the latest SOTA image model from OpenAI, delivering powerful and high-quality capabilities for creating, generating, and editing images. Experience it right now on Commonstack! 🚀

Big W for the @commonstack_ai community 🔥 GPT Image 2 is finally here. The model lineup on Commonstack keeps getting stronger 🚀 Can't wait to see what everyone builds with it 🎨

图像模型更新得越来越快，但我更在意的不是又多了一个选择 而是文字、图片、编辑，正在被压进同一条工作流里。 GPT Image2上线 @commonstack_ai 后，创作者少切一次工具，构建者就少一层摩擦 下一轮竞争，拼的不是模型数量，而是谁能让创意更快落地🎨 #Commonstack

GPT Image 2 is available now on @commonstack_ai This is the latest SOTA image model from OpenAI for you to create, generate or edit images. Try now on Commonstack 🎨

🔹Flexibility & Control: Customize your model outputs with standard parameters. 🔹Pay-As-You-Go: No monthly subscriptions—pay only for the tokens you use. 🔹Production Reliability: Optimized routing for maximum stability and lower latency. Commonstack: commonstack.ai/

Take full control of your AI inference with CommonStack 🚀 X: @Gradient_HQ @commonstack_ai 1/2 🧶

🚨 [New supply chain attack declared]: @common-stack/generate-plugin @common-stack/generate-plugin is an npm package posing as a plugin generator under the @common-stack scope, blending in as legitimate developer tooling. Malware was found in it (GHSA-6p55-6hvr-3xmg). Any system with it installed or running is fully compromised. → Rotate all secrets and signing keys from a clean machine, remove the package, then audit/rebuild Full details 👇 supplychainattack.org/incide… #supplychain #SupplyChainSecurity #infosec #CyberSecurity #npm #commonstack #malware #DevSecOps #AppSec #ThreatIntel #OpenSource

./ Gradient Commonstack Single integration, many providers. Commonstack provides access to the latest models from major providers without rewriting code or managing credentials. ✦ Pay only for what you use No monthly subscriptions. You're billed per token with consolidated billing across all providers. 👉 Commonstack: commonstack.ai One API key, two protocol endpoints Access to models from OpenAI, Anthropic, Google, DeepSeek, MiniMax, Zhipu, xAI, and more. @Gradient_HQ @commonstack_ai

A lot of people have asked me recently whether I have left @Gradient_HQ, and some have also been curious about why. I’ve even seen a few unnecessary speculations. Yes, I have left Gradient, and joined @vidmuse_ai . There is no dramatic story behind this decision. I left simply because I want to keep challenging myself as much as possible before the AI revolution becomes more settled and consolidated — whether that happens around 2029, 2030, or later. Video agent is, in my view, still one of the areas with the most uncertainty and possibility. I also believe that video may become one of the most important forms through which world models are built and experienced over the next five years. Recently, many people have been discussing what world models really are. I have just chosen my answer. For me, Gradient was a very important chapter in my years of working in startups. I loved it deeply. I contributed almost every resource I could. I introduced my own friends to the project. I paid out of my own pocket for some activities. I poured a huge amount of energy into it. Even during my trip to Japan, where I proposed to my girlfriend, I was still thinking about my responsibilities there. I am especially proud of the community team we built. At one point, we had an 8% Discord online rate — a number that could beat 99% of crypto projects, and even many Web2 AI communities. Our community activities were creative, interesting, and full of energy. Even today, there are still loyal community members patiently explaining and educating others about our distributed inference engine, Parallax; our decentralized reinforcement learning framework, Echo; and our aggregated model API platform, CommonStack. Every core community member was amazing. We solved many difficult problems together, and I learned so much from them. I sincerely hope the sky above Gradient remains clear, bright, and pure. Life is only about 30,000 days long. So we have to keep moving closer to what we truly want in our hearts and souls. I hope all of you can do the same. My own journey has always been full of choices — and of being chosen. Games, crypto, AI — each of these fields has shown me beautiful landscapes. But deep down, I know what I truly want: to build something meaningful, to influence people in a positive way, and to create real value. At the end, let me share with you what Brad Pitt said. I don’t know if any of what we do matters in the end. What I do know that matters is the people we hitch ourselves to, and the indelible mark they leave upon our very being.

Claude Opus 4.8 has discovered a serious vulnerability in ZEC, causing this token's price to drop sharply in recent days. Now you can use Claude Opus 4.8 directly on Commonstack, experience its power today. @commonstack_ai

◼️57 AI Models, 11 Providers, One Platform. Most people are still debating: "Which AI model is the best?" But the real question is: Why limit yourself to just one? ◽CommonStack gives access to 57 AI models from 11 providers in a single dashboard. No switching platforms. No managing multiple accounts. Just pick the model that fits the task. ◽Need premium reasoning? > Claude Opus 4.8 > GPT 5.4 > Gemini 2.5 Pro Perfect for: research strategy complex problem solving ◽Need speed lower costs? > DeepSeek V4 Flash > Gemini 3.5 Flash > Gemini 2.5 Flash Great for: summaries chatbots automation high-volume workloads ◽Need the cheapest options? > DeepSeek V4 Flash > Gemini Flash Lite Small cost. Massive scale. Not every task needs a premium model. ◽One thing I like: You can compare models by: > Price > Context Window >Provider > Performance All from the same interface. ◽The AI future won't be: "One model for everything." It will be: "Different models for different jobs." And platforms like CommonStack make that possible. @Gradient_HQ @commonstack_ai

Excited to break down Commonstack the unified access layer that’s making frontier AI instantly usable and insanely efficient inside the @Gradient_HQ ecosystem. Commonstack is the go to platform where you get 50 frontier models from Anthropic, Google, OpenAI, xAI and more all under one roof with a single API key. No juggling multiple accounts, no hunting for credits, no vendor lock in. Just plug in and run the absolute latest releases the moment they drop. Claude Opus 4.8, Gemini 3.5 Flash, and everything else is already live and ready with test credits at zero cost so you can start experimenting immediately. But the real game changer is UncommonRoute their fully open source automatic LLM router that’s now available for everyone. It runs 100 percent locally on your machine as a drop in OpenAI compatible proxy. No code changes needed for Claude Code, Cursor, Codex, or any OpenAI SDK workflow. Here’s how the tech actually works: for every single request or agent step, UncommonRoute looks at three local signals conversation metadata, embeddings via a lightweight BGE classifier, and structural complexity then instantly routes the prompt to the cheapest model that can still deliver top tier quality. Simple tasks go to fast cheap models, complex reasoning or coding jumps to the heavy hitters. The decision is explainable, visible in a clean local dashboard, and you can tweak policies on the fly with modes like auto, fast, or best. The numbers are wild. On real benchmarks like SWE bench it delivers 53 percent lower API costs while keeping accuracy basically identical. Overall it’s hitting 82 percent cost savings and 79.4 percent routing accuracy across thousands of tasks. And because everything stays local you keep full privacy, full control, and you can even give thumbs up or thumbs down feedback to train a personal overlay model that gets smarter over time without touching the base classifier. This is exactly the kind of stack the agentic future needs. When every step in a long running agent loop can cost real money, intelligent per step routing turns expensive inference into something sustainable. Pair it with Gradient’s Parallax distributed inference layer and Lattica’s peer to peer mesh and you’re not just accessing models you’re owning the entire pipeline from edge to global scale. Hexx and the Gradient crew have been pushing this hard because it’s the practical bridge between frontier closed models and the fully open decentralized infrastructure they’re building. Inference costs are skyrocketing with agentic adoption, yet Commonstack UncommonRoute flips the script and gives you more intelligence per dollar than ever before. If you’re building agents, coding daily, or just want to run the best models without burning cash, Commonstack is the common stack you’ve been waiting for. Open source router is live on GitHub right now, the platform is at commonstack.ai with signup credits ready, and the whole thing slots perfectly into the open intelligence flywheel @Gradient_HQ is accelerating. The agentic era doesn’t have to be expensive. It just has to be smart. Commonstack makes it both. @Gradient_HQ @davidz9 @HexxRL @yuangao @gradientintern

./ Commonstack Team Accounts Did you know that you can create a shared account for your team on Commonstack? If you're a builder, this is a great opportunity for you You can create a team from your account settings, invite members, and share a common balance across all team members. What can you do with Team Accounts? ✦ Invite teammates to collaborate under a single account ✦ Share one balance and billing system across the entire team ✦ Create API keys with custom usage limits for different projects ✦ Easily switch between your personal account and team workspace Create a team directly from your account settings and start collaborating with others in a shared workspace. Once the team is created, you can easily switch between your personal account and your team account whenever you log in. ---- If you have a team and actively use AI, this is a great way to conveniently manage your team members' accounts and keep track of expenses. Build with Gradient @Gradient_HQ @commonstack_ai

CommonStack 🔥 One API. 40 beasts. Zero drama. Smart routes your prompts to the fastest & cheapest model _ Claude, DeepSeek, Qwen, whatever cooks best. Pay-as-you-go, no key chaos. Just plug in and go nuclear. ⚡ CommonStack = AI without the headache. @commonstack_ai

谷歌的 Gemini 3.5 Flash 现已在 @commonstack_ai 上线了 commonstack_ai是CommonStack（由Gradient公司开发）的一个AI模型聚合平台 提供统一的OpenAI兼容API，让开发者和企业能够便捷访问包括Google Gemini 3.5 Flash在内的55 主流大语言模型 1. 一站式模型访问模式：通过单一API连接多家顶级AI提供商，无需管理多个API密钥和集成不同接口 2. 统一接口标准：支持标准参数，确保应用行为一致，简化开发与切换成本 3. 灵活控制：提供模型输出与操作设置的自定义工具，适配不同场景需求 4. 按使用付费：无订阅费，仅按处理的token数量计费，多提供商统一账单 5.生产级可靠：优化路由与稳定性，提供比单独管理连接更好的延迟与可靠性 该模型主打代理任务和编码性能，以极低的成本匹配或超越许多其他模型，特别适合需要速度、长上下文和可靠工具使用的生产工作负载，如代理循环、终端自动化、多文件编码和多模态文档分析等场景 通过CommonStack官网访问可获得免费的token体验额度

周五是什么日子，为了周末更好的休息，必须努力的日子！ 当大厂把AI当成自家禁脔，Gradient 偏要把前沿技术拽到普通人手里。 它的杀手锏是智能路由——简单活儿交给便宜模型，复杂任务才请出顶尖大模型，像精打细算的管家，绝不让你多花一分冤枉钱。开源工具 UncommonRoute 更是开箱即用，个人开发者装完就能跑。 Commonstack 平台更像一个免费试玩区，Gemini 3.5 Flash、Claude Opus 4.8 这些顶尖模型，不用砸钱就能上手折腾。至于算力焦虑？Gradient 用分布式网络把千家万户的闲置电脑攒成算力池，彻底打破云巨头的垄断。 说到底，Gradient 不是在卖概念，而是在干一件实在事：把高高在上的 AI，从云端拽到你的桌面上。未来比拼的不是谁家模型参数更吓人，而是谁能在每一次调用时都做出最聪明的选择。 开放、普惠、接地气——这才是 AI 该有的样子。 @Gradient_HQ @HexxRL

TwinRouterBench 是一项非常及时且务实的开源贡献。它精准抓住了 Agentic Era 的核心痛点——从单一 prompt 路由转向真实多步、长时序 Agent 工作流中的 per-step routing。 通过 Static Dynamic 双轨设计，它不仅为路由器提供了高质量的训练数据，还用真实 SWE-bench 验证了cheapest sufficient model策略的有效性，在成本降低 53% 的同时保持了接近的解决率。这为整个行业提供了一个可量化的基准，有望推动更多团队从盲目用顶级模型转向智能性价比路由。 在成本压力日益增加的当下，Commonstack 做的这件事既有学术价值，也有很强的工程落地意义。值得关注。 @Gradient_HQ