Hyperbolic’s cover photo
Hyperbolic

Hyperbolic

Software Development

San Francisco , CA 3,760 followers

High-performance inference & on-demand GPU clusters for teams that refuse to overpay

About us

Hyperbolic provides high‑performance GPU clusters and managed inference for AI startups and ML teams that need reliable capacity without hyperscaler pricing. Teams cut their AI infrastructure costs by using Hyperbolic's clusters optimized for distributed inference, model training and finetuning Researchers use Hyperbolic to launch new models faster, avoid GPU waitlists, and scale from prototype to production on the same platform.

Website
https://hyperbolic.ai/
Industry
Software Development
Company size
11-50 employees
Headquarters
San Francisco , CA
Type
Privately Held

Locations

Employees at Hyperbolic

Updates

  • Not every model needs to save lives, but when one does, we get excited. One foundation model that pushes the state-of-the-art across 300+ pathologies and leads to superhuman breakthroughs (i.e. predicting future cancer) that generalize across continents? Now that’s insane. Massive respect to Adam Yala and team!

    View profile for Adam Yala

    Co-founder and CEO, Voio Inc; Assistant Professor of Computational Precision Health at UC Berkeley and UCSF.

    Today, we’re excited to introduce Voio, a frontier AI lab dedicated to healthcare. As a computer scientist, I’ve seen first hand how the frontier of AI innovation can make it more fun and more empowering to do my job; every year is a better year to be an AI researcher. In nearly a decade of working in healthcare, I’ve never seen a physician feel the same way; the tools stay the same, the workload gets higher, and life gets just a little bit harder. The patient experience rarely inspires delight. At Voio, we’re building a frontier lab to bring the most ambitious AI innovation to where it matters most, healthcare. Starting in radiology, we’re building a unified reading platform that supports radiologists across every scan and modality. Our mission at Voio is to create technology that helps radiologists work faster without sacrificing accuracy—restoring time for interpretation, reducing burnout, and enabling more proactive, preventative care. There is an incredibly exciting opportunity to make health care that anticipates both physician and patient needs. As part of this launch, we’re releasing Pillar-0, our foundation model for radiology developed by my lab at UC Berkeley and UCSF. Pillar-0 processes full CT/MRI volumes, delivers 10–17 point AUC gains over Google, Microsoft, and Alibaba’s best public models across 350+ findings. Pillar-0 raises the bar for radiology foundation models—better, faster, more data efficient, and it’s already improving state-of-the-art in clinical frontier tasks like predicting future lung cancer. Pillar-0 is open-source, and you can build on it today. I can’t wait to see what it enables. Releasing Pillar-0 in the open is part of our broader vision for Voio. Healthcare demands performance transparency, rigour, and collaboration. When the community can’t independently test models, the whole community suffers. We’re excited to continue working with the academic partners to advance the science of precision health. I’m excited for this milestone and can’t wait to share even bigger results soon. Onwards! Website: Voio.com Full company announcement here: https://lnkd.in/d_wr846C Full Pillar-0 announcement here:  https://lnkd.in/dbZ9dBps Pillar-0 code, models, documentation, and evaluation tools are available here: https://lnkd.in/d7XnDNrU

  • View organization page for Hyperbolic

    3,760 followers

    Hyperbolic teamed with NVIDIA to provide Day 0 support for the new Nemotron Models We're excited to announce that Hyperbolic now hosts the latest NVIDIA Nemotron models, a powerful family of open models, datasets, and techniques designed to help enterprises build high-accuracy, specialized agentic AI. Why Nemotron? Most enterprises don't need another giant foundation model. They need specialized agents that understand their documents, dashboards, videos, and workflows—and that can act: retrieve facts, fill forms, reconcile data, route tickets, and follow safety policies. Nemotron delivers with: ✅ Open weights, open data, open recipes for auditability and trust in regulated environments ✅ Top-tier accuracy & efficiency for vision, reasoning, and agentic tasks ✅ Run-anywhere packaging (including NVIDIA NIM) optimized for hybrid and sovereign AI deployments Hyperbolic's multi-cloud GPU infrastructure and inference platform brings Nemotron to life with: - Lower latency and cost for production workloads - Full transparency into data and training processes - On-demand and reserved NVIDIA Hopper and Blackwell GPUs (H100s, H200s) at industry-leading prices - Production SLAs and enterprise controls Featured Models: 🧠 Nemotron Nano 2 VL (12B) A compact hybrid Transformer-Mamba VLM delivering up to 10× higher token throughput with state-of-the-art accuracy across OCR, charts, video understanding, and document intelligence. Supports 128k context and multi-image/video inputs. 📄 Nemotron Parse 1.1 (~1B) A lightweight VLM built to extract structured information from PDFs, contracts, research reports, charts, and diagrams. Get Started Today: Try out Nemotron Nano 2 VL on Hyperbolic's inference platform: https://lnkd.in/geq9NSkx Read the full blog: https://lnkd.in/gFz9r-n4

    • No alternative text description for this image
  • Ready to Compete for Compute? ♠️💻 Join us October 28th in San Francisco for Poker Night [Compute Edition], hosted by Hyperbolic, Entrepreneurs First and The AI Collective No buy-ins. No stakes. Just (free) compute and fun. 🍕 6:30 PM — Food + networking ♣️ 7:00 PM — Poker showdown: everyone starts with $compute credits in chips and can walk out with more. Meet the founders, engineers, and researchers building the next frontier of AI infrastructure. Spots are limited, registration is approval-only. 🎟 Apply to join → https://luma.com/ffssgi8y Come for the fun, stay for the compute!

    • No alternative text description for this image
  • According to Artificial Analysis, Hyperbolic stands out with strong performance across context lengths. Our gpt-oss-120b model delivers 402 tokens/sec and maintains consistent throughput across both short and long contexts. ⚡ Consistency: Whether it’s 1k or 100k tokens, our TPS barely dips. 💰 Efficiency: High throughput at low cost means real-world savings. 🏭 Production Ready: Designed for developers who need reliability at scale—not just benchmarks. In practice, this means teams can train, fine-tune, and deploy workloads with confidence, knowing performance won’t collapse as context sizes grow. At Hyperbolic, we’re not just aiming for speed, we’re building the most reliable inference layer for real production workloads. 👉 If you’re scaling LLM apps and care about consistent performance, predictable costs, and production-grade reliability, we’d love to connect! Try it out at https://lnkd.in/gQgGf3gx

    • No alternative text description for this image
  • 🚀 NVIDIA-Powered Qwen3-Next: Now Live at Hyperbolic Hyperbolic's superior benchmarks for Qwen3-Next-80B-A3B are powered by our NVIDIA clusters, featuring on-demand H100 and H200 GPUs optimized for AI workloads. NVIDIA's H100 and H200 GPUs power our infrastructure with their Hopper architecture, providing the compute necessary for Qwen3-Next's massive MoE models. While H100s offer 80GB of HBM3 memory, H200s upgrade to 141GB of HBM3e memory with up to 45% better performance. As the first provider to launch inference for this model in direct partnership with the Qwen team, Hyperbolic delivers the highest performance at the lowest price. Try it out today: Qwen3-Next-80B-A3B-Thinking: https://lnkd.in/g2z6uMjK Qwen3-Next-80B-A3B-Instruct: https://lnkd.in/gtaPGSYp Or get started with Hyperbolic GPUs: https://lnkd.in/g693TdQ3

    • No alternative text description for this image
  • Qwen3-Next-80B-A3B Benchmarking: Why Hyperbolic Leads Recent benchmarking from Artificial Analysis puts Hyperbolic at the very top for Qwen3-Next-80B-A3B-Thinking and Qwen3-Next-80B-A3B-Instruct — across every key metric: speed, latency, and cost. 💡 Hyperbolic was also the first to launch inference for this model, working directly with the Qwen team. Model: Qwen3-Next-80B-A3B-Instruct 👩🏻💻 > Output Speed: 237 tokens/sec > Latency (TTFT): 0.52s > Blended Price: $0.30 / 1M tokens Model: Qwen3-Next-80B-A3B-Thinking 💭 > Output Speed: 288 tokens/sec > Latency (TTFT): 0.55s > Blended Price: $0.30 / 1M tokens Hyperbolic’s industry-leading performance is built on on-demand NVIDIA H100s and H200s, available for rent. These GPUs power the speed and efficiency that set us apart in the benchmarks. 👉 Try it now: https://lnkd.in/g693TdQ3

    • No alternative text description for this image
  • View organization page for Hyperbolic

    3,760 followers

    🚀 Qwen3-Next is now live at Hyperbolic in BF16! This marks a major leap in efficiency for open-source AI: 🔹 80B MoE with only 3B active params 🔹 10× cheaper to train vs. Qwen3-32B 🔹 10× inference throughput on >32K tokens Proud to partner with Alibaba Qwen! 🙌 , huge kudos to the team for continuing to push the frontier of open-source AI. We’re also the first to serve Qwen3-Next on Hugging Face 🤗 Hyperbolic’s industry-leading performance is built on on-demand NVIDIA H100s and H200s, available for rent. These GPUs power the speed and efficiency that set us apart in the benchmarks. Try it out now: - Qwen3-Next-80B-A3B-Thinking: https://lnkd.in/g9sKcUtd - Qwen3-Next-80B-A3B-Instruct: https://lnkd.in/gBRTdK3y - Hugging Face: https://lnkd.in/g85qAm_q

Similar pages

Browse jobs

Funding