Together AI

🧪 New Notebook Drop: Evaluating LLMs for harmful outputs! Which models are actually safe for prod? We built an LLM-as-a-Judge pipeline using the Together Evals API to compare models on harmfulness. Code👇

💡Stop guessing. Start benchmarking. Every team building with LLMs runs into the same problems: “Which model is actually better for my task?” “Can I trust this before I ship it?” “How do I catch errors before users do?” Together Evaluations solves these problems — fast. This early preview of our new evaluation tool lets you define task-specific benchmarks and use a strong LLM as a judge to: ✅ Compare models side-by-side ✅ Score responses against your own criteria ✅ Classify outputs into custom labels — from safety to sentiment You can evaluate any serverless model on Together AI today. Later this summer, you’ll be able to evaluate fine-tuned models, custom models, and even commercial APIs — all in one place. 📊 Use it to test prompts, validate new use cases, find the best open-source model for your task. Learn more (links in comments!)

🧠 𝐓𝐡𝐫𝐞𝐞 𝐦𝐨𝐝𝐞𝐥𝐬, 𝐨𝐧𝐞 𝐰𝐞𝐞𝐤. 𝐐𝐰𝐞𝐧𝟑-𝟐𝟑𝟓𝐁-𝐀𝟐𝟐𝐁-𝐓𝐡𝐢𝐧𝐤𝐢𝐧𝐠 𝐣𝐮𝐬𝐭 𝐡𝐢𝐭 𝐓𝐨𝐠𝐞𝐭𝐡𝐞𝐫 𝐀𝐈 🔥 This 235B parameter reasoning model just set new standards for open-source thinking models. 92.3 on AIME25 (vs 81.5 from the previous version) with transparent reasoning chains you can follow through complex problems. 𝐖𝐡𝐲 𝐭𝐡𝐢𝐬 𝐦𝐚𝐭𝐭𝐞𝐫𝐬 𝐟𝐨𝐫 𝐩𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐨𝐧: ⚡ State-of-the-art open-source reasoning 🧠 256K context — 𝘆𝗼𝘂𝗿 𝗲𝗻𝘁𝗶𝗿𝗲 𝗰𝗼𝗱𝗲𝗯𝗮𝘀𝗲 𝗶𝗻 𝘄𝗼𝗿𝗸𝗶𝗻𝗴 𝗺𝗲𝗺𝗼𝗿𝘆 🔧 Transparent debugging — trace through complex logic step-by-step 🎯 235B total, 22B activated — 𝗳𝗿𝗼𝗻𝘁𝗶𝗲𝗿 𝗽𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲, 𝗠𝗼𝗘 𝗲𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝗰𝘆 Perfect for autonomous debugging, mathematical proofs, and complex reasoning where you need both breakthrough performance and complete transparency. Three frontier Qwen models in one week on Together AI. When reasoning transparency meets our production-scale infrastructure, you get AI that works where traditional models hit walls. 𝐒𝐭𝐨𝐩 𝐜𝐨𝐦𝐩𝐫𝐨𝐦𝐢𝐬𝐢𝐧𝐠 𝐨𝐧 𝐫𝐞𝐚𝐬𝐨𝐧𝐢𝐧𝐠 𝐪𝐮𝐚𝐥𝐢𝐭𝐲. 𝐒𝐭𝐚𝐫𝐭 𝐝𝐞𝐩𝐥𝐨𝐲𝐢𝐧𝐠 𝐚𝐭 𝐬𝐜𝐚𝐥𝐞. 🛠️

🤖 𝐐𝐰𝐞𝐧𝟑-𝐂𝐨𝐝𝐞𝐫 𝐣𝐮𝐬𝐭 𝐝𝐫𝐨𝐩𝐩𝐞𝐝 𝐨𝐧 𝐓𝐨𝐠𝐞𝐭𝐡𝐞𝐫 𝐀𝐈. 🔥 This 480B agentic coding model isn't just writing functions - it's executing the kind of complex, repository-scale engineering work that breaks most AI assistants. 💻 𝐖𝐡𝐲 𝐭𝐡𝐢𝐬 𝐜𝐡𝐚𝐧𝐠𝐞𝐬 𝐡𝐨𝐰 𝐲𝐨𝐮 𝐛𝐮𝐢𝐥𝐝 𝐬𝐨𝐟𝐭𝐰𝐚𝐫𝐞: ⚡ Autonomous debugging across microservices 🏗️ Legacy system modernization that actually works 🧠 Repository-wide context understanding (256K tokens) 🔧 Multi-step engineering workflows without hand-holding 📈 Claude Sonnet 4-level performance on complex tasks Your entire codebase fits in its working memory while it plans migrations, traces bugs, and implements features that span multiple services. The kind of engineering work that used to take weeks. 🚀 Together AI's infrastructure makes deploying this 480B monster as simple as an API call - with the reliability and performance optimization that turns frontier AI into production reality. ⚙️ 𝐒𝐭𝐨𝐩 𝐰𝐫𝐢𝐭𝐢𝐧𝐠 𝐟𝐮𝐧𝐜𝐭𝐢𝐨𝐧𝐬. 𝐒𝐭𝐚𝐫𝐭 𝐛𝐮𝐢𝐥𝐝𝐢𝐧𝐠 𝐬𝐲𝐬𝐭𝐞𝐦𝐬. 🛠️

🤖 𝐐𝐰𝐞𝐧𝟑-𝐂𝐨𝐝𝐞𝐫 𝐣𝐮𝐬𝐭 𝐝𝐫𝐨𝐩𝐩𝐞𝐝 𝐨𝐧 𝐓𝐨𝐠𝐞𝐭𝐡𝐞𝐫 𝐀𝐈. 🔥 This 480B agentic coding model isn't just writing functions - it's executing the kind of complex, repository-scale engineering work that breaks most AI assistants. 💻 𝐖𝐡𝐲 𝐭𝐡𝐢𝐬 𝐜𝐡𝐚𝐧𝐠𝐞𝐬 𝐡𝐨𝐰 𝐲𝐨𝐮 𝐛𝐮𝐢𝐥𝐝 𝐬𝐨𝐟𝐭𝐰𝐚𝐫𝐞: ⚡ Autonomous debugging across microservices 🏗️ Legacy system modernization that actually works 🧠 Repository-wide context understanding (256K tokens) 🔧 Multi-step engineering workflows without hand-holding 📈 Claude Sonnet 4-level performance on complex tasks Your entire codebase fits in its working memory while it plans migrations, traces bugs, and implements features that span multiple services. The kind of engineering work that used to take weeks. 🚀 Together AI's infrastructure makes deploying this 480B monster as simple as an API call - with the reliability and performance optimization that turns frontier AI into production reality. ⚙️ 𝐒𝐭𝐨𝐩 𝐰𝐫𝐢𝐭𝐢𝐧𝐠 𝐟𝐮𝐧𝐜𝐭𝐢𝐨𝐧𝐬. 𝐒𝐭𝐚𝐫𝐭 𝐛𝐮𝐢𝐥𝐝𝐢𝐧𝐠 𝐬𝐲𝐬𝐭𝐞𝐦𝐬. 🛠️

🎉 Congratulations to our partner 5C on securing $835 million in capital from Brookfield and Deutsche Bank! At Together AI, we’ve long believed that frontier AI demands an entirely new infrastructure stack—one purpose-built from the ground up. That means rethinking everything: data centers, hardware, software, and services. We’re proud to collaborate with 5C as we continue building and scaling Frontier AI Factories: new infrastructure purpose-built for next-generation AI, powering the training and deployment of the most advanced models and agents. Here’s to accelerating the future of AI—together. 🚀

Deep dive into RAG architecture from our course with Andrew Ng and Zain Hasan 🧠 This 2-minute snippet covers the foundational patterns for building retrieval-augmented systems.

🧠 𝐐𝐰𝐞𝐧𝟑 𝐣𝐮𝐬𝐭 𝐥𝐞𝐯𝐞𝐥𝐞𝐝 𝐮𝐩 𝐨𝐧 𝐓𝐨𝐠𝐞𝐭𝐡𝐞𝐫 𝐀𝐈. 🚀 The new Qwen3-235B-A22B-Instruct-2507-FP8 isn't just another model update - it's a meaningful leap forward in how AI handles complex reasoning, coding, and multilingual tasks. 📈 𝐖𝐡𝐚𝐭 𝐦𝐚𝐤𝐞𝐬 𝐭𝐡𝐢𝐬 𝐮𝐩𝐠𝐫𝐚𝐝𝐞 𝐰𝐨𝐫𝐭𝐡 𝐲𝐨𝐮𝐫 𝐚𝐭𝐭𝐞𝐧𝐭𝐢𝐨𝐧: 📊 → Smarter reasoning and math capabilities 🧮 → Better coding and tool integration 🤖 → Better alignment with user preferences for subjective tasks 🎯 → Works across 100+ languages 🌍 → Handles massive documents (262K context) 📚 Running on Together AI means you get this performance upgrade with our 99.9% uptime and the price-performance optimization that makes frontier models practical at scale. 🏢 𝐔𝐩𝐠𝐫𝐚𝐝𝐞 𝐲𝐨𝐮𝐫 𝐀𝐈 𝐬𝐭𝐚𝐜𝐤 𝐧𝐨𝐰. 🚀

🧠 𝐐𝐰𝐞𝐧𝟑 𝐣𝐮𝐬𝐭 𝐥𝐞𝐯𝐞𝐥𝐞𝐝 𝐮𝐩 𝐨𝐧 𝐓𝐨𝐠𝐞𝐭𝐡𝐞𝐫 𝐀𝐈. 🚀 The new Qwen3-235B-A22B-Instruct-2507-FP8 isn't just another model update - it's a meaningful leap forward in how AI handles complex reasoning, coding, and multilingual tasks. 📈 𝐖𝐡𝐚𝐭 𝐦𝐚𝐤𝐞𝐬 𝐭𝐡𝐢𝐬 𝐮𝐩𝐠𝐫𝐚𝐝𝐞 𝐰𝐨𝐫𝐭𝐡 𝐲𝐨𝐮𝐫 𝐚𝐭𝐭𝐞𝐧𝐭𝐢𝐨𝐧: 📊 → Smarter reasoning and math capabilities 🧮 → Better coding and tool integration 🤖 → Better alignment with user preferences for subjective tasks 🎯 → Works across 100+ languages 🌍 → Handles massive documents (262K context) 📚 Running on Together AI means you get this performance upgrade with our 99.9% uptime and the price-performance optimization that makes frontier models practical at scale. 🏢 𝐔𝐩𝐠𝐫𝐚𝐝𝐞 𝐲𝐨𝐮𝐫 𝐀𝐈 𝐬𝐭𝐚𝐜𝐤 𝐧𝐨𝐰. 🚀

Congrats to Slingshot AI on launching Ash today! 🎉 They’ve been training this psychology foundation model in partnership with us for the last 18 months - purpose-built for therapy rather than adapting general models. Incredible to see it finally live. 👏 Together AI is glad to be the platform of choice for their solution and we are excited to see users explore AI that’s already helping thousands access mental health support.