Together AI

?? AI requires more than just GPUs — it demands high-bandwidth networking, advanced cooling, and robust fault tolerance. Together AI and Hypertec Cloud have partnered to deliver clusters of thousands of GPUs — oftentimes delivering infrastructure in a few weeks what would take other clouds 6+ months. ?? Watch this video featuring Together AI's CEO, Vipul Ved Prakash and Hypertec Cloud's President, Jonathan A., to learn more about how we are collaborating to jointly deploy large-scale GPU clusters accelerated by NVIDIA Blackwell platform and our own Together Kernel Collection.

New Notebook: Search and Planning Agent using LangGraph and Together! Build an agent that plans and executes with web search and tool use. Check out the link below and docs! Notebook: https://lnkd.in/gEWZ3iTT Docs: https://lnkd.in/g7cRSCVs

?? We’re honored to share that Together GPU Clusters have been awarded the ClusterMAX Gold Rating by SemiAnalysis—recognizing our leadership in GPU infrastructure, performance, and technical support. This independent evaluation highlights Together AI’s: ?? Reliable, secure infrastructure ?? Industry-leading GPU kernel optimizations ?? Deep research and engineering & support expertise ?? Leading price-performance with flexible consumption models Together GPU Clusters are powered by the latest NVIDIA HGX B200 and GB200 NVL72 hardware, and supported by expert Field Engineers and our proprietary Together Kernel Collection, delivering 30%+ faster training operations. We're also raising the bar with Instant GPU Clusters, giving teams fast, self-service access to 8–64 NVIDIA GPUs, with transparent pricing and free data transfer. ?? Read the full story: https://lnkd.in/d-r8bGwr ? Try Instant GPU Clusters: together.ai/instant ?? Explore Together GPU Clusters: together.ai/gpu-clusters

?? "The Open-Source AI Acceleration Cloud." That’s the headline of the latest issue of The GPU Newsletter — and it’s all about Together AI. We’re honored to be featured for our work building a faster, more transparent, more open alternative to the traditional cloud. No lock-in. No hidden fees. Just world-class AI infrastructure, purpose-built for developers and researchers. ?? Rolled out Instant GPU Clusters, offering self-service deployments of up to 64 GPUs ?? Joined the NVIDIA Cloud Partner Network, gaining early access to GB200 NVL72s and deploying a 36,000-GPU cluster ?? Raised $305M in Series B funding from top-tier investors including General Catalyst, Salesforce Ventures, and NVIDIA ?? Launched Reasoning Clusters and ranked among the fastest API providers for DeepSeek-R1 Huge thanks to Ben Baldieri for capturing what we’re building and why it matters ??

New Notebook: Agentic RAG with DeepSeek-R1 using LangGraph and Together! Build an agent that decides when to retrieve from a vector DB, evaluates result quality and generates. Reasoning models are great at identifying irrelevant retrieved chunks and ignoring them! Notebook:?https://lnkd.in/ggcyZw-f Docs:?https://lnkd.in/g7cRSCVs

DeepSeek-V3-0324 is now live on chat.together.ai!

?? New Model Drop: DeepSeek-V3-0324 is now live on Together AI ?? ?? Try it at https://lnkd.in/gEsyjcEH DeepSeek-V3-0324 delivers major improvements over its predecessor across multiple challenging benchmarks. This latest release showcases the rapid evolution of open-source models—and it’s available now on the Together AI platform. ?? Performance gains vs. DeepSeek-V3: MMLU-Pro (expert-level reasoning across multiple subjects) ??: 75.9 → 81.2 (+5.3) GPQA Diamond (graduate-level science and commonsense questions) ??: 59.1 → 68.4 (+9.3) AIME 2024 (mathematical olympiad-style problem solving) ??: 39.6 → 59.4 (+19.8) LiveCodeBench (code generation and reasoning) ??: 39.2 → 49.2 (+10.0) Try DeepSeek-V3-0324 on the Together Playground ?? https://lnkd.in/g8ma_B5a Together AI makes it easy to explore cutting-edge models with fast, cost-effective inference—whether you're building apps or just experimenting.

??? QwQ-32B – Alibaba's latest reasoning-optimized large language model – is available on Together AI! ?? https://lnkd.in/gEcNgss2 ?? What is QwQ-32B? A 32-billion-parameter model trained using Reinforcement Learning (RL) techniques to significantly enhance mathematical reasoning, coding proficiency, and problem-solving. ?? Why is this important? QwQ-32B exemplifies remarkable efficiency by delivering performance on par with much larger models. Notably: - Benchmark Performance: It achieves 65% on the GPQA (Graduate-level Physics Question Answering) benchmark, 50% on AIME (American Invitational Mathematics Examination), over 90% on MATH-500, and 50% on LiveCodeBench, showcasing its advanced reasoning and coding capabilities. - Parameter Efficiency: Despite having 32 billion parameters, QwQ-32B matches the capabilities of models like DeepSeek's R1, which has 671 billion parameters (with 37 billion activated). This efficiency translates to reduced computational requirements and cost-effective deployment, making advanced AI more accessible. ?? How was it trained? The model's unique reinforcement learning pipeline included: - Fine-tuning with accuracy verifiers and code execution to refine problem-solving. - Instruction-tuned RL to align with human preferences and improve general usability. ?? What is QwQ-32B good for? ?? Advanced math problem solving ?? High-quality code generation ?? Complex reasoning and structured tasks ?? Secure and Private Deployment We understand the importance of security and data privacy when working with AI models. QwQ-32B on Together AI is designed with enterprise-grade security measures, ensuring that sensitive data remains protected. Users can deploy the model in secure environments with fine-grained access control and opt out of data sharing, keeping all interactions private. With QwQ-32B now running on Together AI, developers and enterprises can leverage this powerful model on our high-performance inference infrastructure while maintaining full control over their data. Put its reasoning to the test, for just $1.20 per million input/output tokens! ?? Try it now on the Together Playground https://lnkd.in/gzhqX4Qp ?? Learn more about the model https://lnkd.in/gEcNgss2

Together and Composio are partnering to bring an easy way for developers to build agents! ?? Use the top open source LLMs on Together AI along with 250+ tools from Composio ?? Your LLMs on Together can now use everything from Gmail to Github as a part of your agents ?? Install in one library: pip install composio-togetherai Want to see it in action? Check out our notebook to see how you can automate sending email with LLMs: https://lnkd.in/gjTVgwkp Learn more in our docs: https://lnkd.in/gfRsgPev

Announcing chat.together.ai – our new consumer chat app! Use ?? DeepSeek R1 (hosted in North America) & other top open source models like ?? Llama to do web search, coding, image generation, & image analysis — all for free. ?? Try it now: https://chat.together.ai/ ?? Blog with use cases: https://lnkd.in/ePmtGvaw We built this to give everyone an easy way to use top open source models in a great UI for free — securely hosted in data centers across the US and Canada. Here are the top use cases: ?? Chat with DeepSeek R1 and other top OSS models ?? Do web search with DeepSeek R1 and other top OSS models ?? Generate code with Qwen Coder 32B ??? Generate images with Flux Schnell ?? Analyze images with Qwen 2.5 VL Let us know what you think! Lots more updates coming.