Together AI

Building powerful AI agents requires realistic evaluations that mimic the unpredictable, dynamic nature of real-world human interactions. Introducing TraitMix, Collinear AI’s simulation product empowering teams to generate persona-driven AI agent interactions. Seamlessly plug these interactions into your workflows and evaluate their effectiveness with Together Evals. Using TraitMix, you can:  ⚪ Select traits & personas ⚪ Choose domains  ⚪ Simulate dialogues  ⚪ Judge automatically Close the loop between interaction, evaluation, and improvement, all in a single ecosystem with Collinear TraitMix + Together Evals.

The future of AI is here – and it’s at #NVIDIAGTC. Yesterday, we showcased live model demos, a few in-booth activities, a surprise cocktail activation, and a theater talk from Together AI’s SVP of Infrastructure Engineering, Mahadev Konar. The conversations were energizing and we hope you enjoyed the putting green in between demos. Today, Mahadev Konar returns to the stage joined by NVIDIA moderator Renee Y., and other tech leaders, to discuss the era of AI Factories—data centers purpose-built for frontier AI. There’s more to come, so join us for the final day of #NVIDIAGTC and visit booth 223!

We’re joining forces with 5C to power the next generation of AI-native development! Launching early 2026, the Frontier AI Factory in Memphis represents a significant milestone in delivering a full stack platform for AI natives to build and scale the next generation of apps. ➡️ Powered by NVIDIA, Dell Technologies, VAST Data, and Together Kernel Collection.

Exciting News: NVIDIA Nemotron-Nano-9B-v2 Now Available on Together AI Open-source NVIDIA Nemotron-Nano-9B-v2 is now accessible through Together AI as both a serverless endpoint and dedicated endpoint! What makes this model special: 🧠 Unified reasoning & chat capabilities: High-quality responses for complex queries 📈 Impressive benchmarks: 97.8% on MATH500, 72.1% on AIME25 🌏 Multilingual support: English, German, Spanish, French, Italian, and Japanese ⚖️ Runtime control: Adjust the "thinking" token budget to optimize accuracy and speed 🔓 Available for commercial use under the NVIDIA Open Model License Whether you're building AI agents, chatbots, RAG systems, or other AI-powered applications, this model offers exceptional reasoning capabilities in a small 9B package!

We’re excited to partner with 5C to build a new Frontier AI Factory in Memphis, launching in early 2026. This facility represents a significant milestone in delivering a full stack platform for AI natives to build and scale the next generation of apps. Powered by NVIDIA, Dell Technologies, VAST Data, and Together Kernel Collection, this purpose-built facility will feature NVIDIA GB200 and is designed to enable the world’s most complex workloads. Together, we are committed to supporting the most demanding applications and empowering the next generation of AI-native development.

Do reasoning models really follow our instructions? 🧠 Together AI’s newest paper "ReasonIF: Large Reasoning Models Fail to Follow Instructions During Reasoning", studies how well large reasoning models (LRMs) follow user instructions during reasoning. TL;DR 👇 Together introduces ReasonIF benchmark and finds: ⚪️Less than 25% of LRMs follow instructions ⚪️Degree of question difficulty impacts compliance Quick fixes to include: ⚪️Multi-turn reasoning  ⚪️Reasoning instruction fine-tuning (RIF) Learn more about our AI Agent team’s latest study: https://bit.ly/42UBSFq

The future of AI is here — and it’s at #NVIDIAGTC! Join us for sessions featuring leaders across Together AI, including VP of Strategic Partnerships Arielle Fidel and SVP of Infrastructure Engineering, Mahadev Konar. We'll dive into how purpose-built cloud infrastructure accelerates AI innovation — and how Together powers the next generation of AI-native pioneers. Kick off the conference with us at our cocktail reception: https://luma.com/mgv7hfw0 Plus we have a few can’t-miss activations at our booth #223, so we'll see you there!

🎬 𝐕𝐢𝐝𝐞𝐨 𝐠𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐨𝐧 𝐢𝐬 𝐧𝐨𝐰 𝐚𝐯𝐚𝐢𝐥𝐚𝐛𝐥𝐞 𝐨𝐧 𝐓𝐨𝐠𝐞𝐭𝐡𝐞𝐫 𝐀𝐈 If you're building multimodal AI apps, you know the headache: one provider for text, another for images, a third for video. Different auth systems, separate rate limits, fragmented billing. That complexity kills velocity. Through our partnership with Runware, we're integrating 20+ video models and 15+ image models alongside the LLMs you're already running - all accessible through the same OpenAI-compatible APIs you already use for text inference. 𝐖𝐡𝐚𝐭 𝐭𝐡𝐢𝐬 𝐦𝐞𝐚𝐧𝐬 𝐟𝐨𝐫 𝐩𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐨𝐧: 🎮 Gaming studios generating environmental assets from gameplay data 🛍️ E-commerce platforms creating personalized product videos based on user preferences 🧠 EdTech apps building custom visual explanations in real-time All through a single platform with unified authentication and billing. No new SDKs to learn. No vendor coordination overhead. 𝐍𝐞𝐰 𝐜𝐚𝐩𝐚𝐛𝐢𝐥𝐢𝐭𝐢𝐞𝐬: ⚡ Video generation with Sora 2 Pro, Google Veo 3.0, Minimax Hailuo, ByteDance Seedance 🎨 Latest image models including Google Imagen 4.0, Gemini Flash, Qwen Image 🔧 Same APIs, same authentication, same billing dashboard you already use 𝐁𝐮𝐢𝐥𝐝 𝐜𝐨𝐦𝐩𝐥𝐞𝐭𝐞 𝐦𝐮𝐥𝐭𝐢𝐦𝐨𝐝𝐚𝐥 𝐰𝐨𝐫𝐤𝐟𝐥𝐨𝐰𝐬 𝐰𝐢𝐭𝐡𝐨𝐮𝐭 𝐬𝐰𝐢𝐭𝐜𝐡𝐢𝐧𝐠 𝐩𝐫𝐨𝐯𝐢𝐝𝐞𝐫𝐬 🛠️