Cerebras

Build AI fastest with Cerebras on AWS Marketplace! We’re live at re:Invent — and Cerebras Inference is now available on AWS Marketplace: https://lnkd.in/gAveM6yi 🚀 Here’s why the most successful AI teams choose Cerebras Inference on AWS: • Fastest real-time inference in the Marketplace • High throughput + ultra-low latency for agents and production workloads • Effortless scaling for concurrency, batch jobs, and mission-critical apps 🏗️ With AWS Marketplace, you get: • A curated, enterprise-ready catalog • Accelerated procurement • Flexible pricing and centralized governance Come see us at #AWSreInvent, we are here all week! Stop by Booth #1772 (Venetian) to see live demos, meet the team, and spin up Cerebras in minutes via Amazon Web Services (AWS) Marketplace. We are proud to be AWS Partners. Meet the team: https://lnkd.in/gKpbdVay

If you're building LLM apps at scale, you’ve probably felt the gap between model quality and model speed. Most systems still struggle with latency, throughput, and cost once real workloads hit. That gap is exactly what we’ll unpack in our upcoming Cerebras x Portkey webinar. Register now! https://luma.com/dzzf3iq8 Portkey is a developer platform that makes it dramatically easier to build real LLM applications. It provides a robust, secure gateway for integrating multiple models — including those powered by the Cerebras Inference API — with built-in routing, observability, guardrails, and production controls. In short: Portkey makes fast models usable. During the session: 1️⃣ Karthik Viswanathan S. and Saurabh Vyas will break down why speed lags quality across the industry, and how Cerebras delivers ultra-high token throughput with a fundamentally different inference architecture. 2️⃣ Vrushank Vyas will show how Portkey turns that performance into real-world wins — from scale and concurrency to agentic workloads.

✨Research like this is what DREAMs are made of...and we’re excited to share it at NeurIPS 2025. ✨ 𝐃𝐑𝐄𝐀𝐌: 𝐃𝐫𝐚𝐟𝐭𝐢𝐧𝐠 𝐰𝐢𝐭𝐡 𝐑𝐞𝐟𝐢𝐧𝐞𝐝 𝐓𝐚𝐫𝐠𝐞𝐭 𝐅𝐞𝐚𝐭𝐮𝐫𝐞𝐬 𝐚𝐧𝐝 𝐄𝐧𝐭𝐫𝐨𝐩𝐲-𝐀𝐝𝐚𝐩𝐭𝐢𝐯𝐞 𝐂𝐫𝐨𝐬𝐬-𝐀𝐭𝐭𝐞𝐧𝐭𝐢𝐨𝐧 𝐅𝐮𝐬𝐢𝐨𝐧 𝐟𝐨𝐫 𝐌𝐮𝐥𝐭𝐢𝐦𝐨𝐝𝐚𝐥 𝐒𝐩𝐞𝐜𝐮𝐥𝐚𝐭𝐢𝐯𝐞 𝐃𝐞𝐜𝐨𝐝𝐢𝐧𝐠 will be presented at the main conference: 📍 Exhibit Hall C, D, E Booth 4009 🗓 Fri, Dec 5 ⏰ 4:30 p.m.–7:30 p.m. PST 𝗙𝗼𝗰𝘂𝘀 𝗼𝗳 𝘁𝗵𝗲 𝗽𝗮𝗽𝗲𝗿: DREAM introduces a speculative decoding framework for vision-language models, combining cross-attention feature injection, entropy-adaptive intermediate feature selection, and visual token compression to accelerate multimodal generation. 𝗧𝗵𝗲 𝘁𝗲𝗮𝗺’𝘀 𝗮𝗽𝗽𝗿𝗼𝗮𝗰𝗵: The team designed a draft–target system that pulls key visual and textual signals from the target model into the draft model, selects the most informative intermediate features based on attention entropy, and compresses visual tokens to significantly reduce latency. 𝗗𝗶𝘀𝗰𝗼𝘃𝗲𝗿𝗶𝗲𝘀: Across LLaVA, Pixtral, SmolVLM, and Gemma3, DREAM delivers up to 3.6× speedup and substantially longer accepted-token spans than prior SD approaches on a broad suite of multimodal benchmarks. 𝗪𝗵𝘆 𝗶𝘁 𝗺𝗮𝘁𝘁𝗲𝗿𝘀: DREAM demonstrates that speculative decoding can be extended effectively to VLMs, enabling faster multimodal reasoning, lower inference cost, and more responsive agentic systems. Grab the code: GitHub: https://lnkd.in/gNDqwFWt Draft model: Hugging Face: https://lnkd.in/grk6zCcr Congratulations to Yunhai Hu, Tianhua X. Rahul Raman, Xingyu Liu, Bo Bao Eric Sather Vithu Thangarasa SAI QIAN ZHANG New York University University of Pennsylvania

Earlier today, The White House announced the launch of the Genesis Mission, a new national initiative to accelerate AI for science. This initiative, led by the U.S. Department of Energy (DOE), will bring the nation's best and brightest together with world leading supercomputers and datasets to radically accelerate transformational work in AI for science and security, health and energy in the nation's interest. An incredible moonshot moment for AI, American innovation, industry, and public-private partnership. Let's go! Full EO: https://lnkd.in/gtnZM4aV Fact sheet: https://lnkd.in/gYZGu_gD Darío Gil | Chris Wright | Hal Finkel | Si Hammond | Thuc Hoang | Brian Spears | Rick Stevens | Michael Papka | Sivasankaran Rajamanickam | James Laros | Sara Prentice | Prasanna Balaprakash | Cerebras | #GenesisMission

🟧 🍴 Our table is set with record-breaking TPS: Turkeys Per Serving. Huge thanks to our partners, customers, and especially the Cerebras team for pushing the limits of what’s possible this year. Have a wafer-scale Thanksgiving! 🎉

🚀 Cerebras is heading to NeurIPS 2025 and we’re bringing an all-star lineup of research, workshops, and community events to San Diego. Here’s where you’ll find us: 📍 NeurIPS Expo Hall Booth 718 Come talk to the team behind the scenes of the world's fastest inference. And don't forget your selfie with WAFER 🟧 ☕ Café Compute: Dec 4 Step into our winter-wonderland coffeeshop for late-night coffee ☕️, donuts 🍩, and snacks 🍿 brought to life by Cerebras, Bain Capital Ventures (BCV), OpenAI, Mercor, and @sfcompute Register: https://lnkd.in/gzQ4ZAJD 📈 8th Neural Scaling Workshop: Dec 5–6 We’re co-organizing two days of talks on frontier training, real-time inference, scaling laws, and the breakthroughs pushing AI forward. https://lnkd.in/gqBjuTN9

🎲 𝐅𝐨𝐫𝐠𝐞𝐭 𝐥𝐮𝐜𝐤. 𝐖𝐚𝐟𝐞𝐫-𝐬𝐜𝐚𝐥𝐞 𝐬𝐩𝐞𝐞𝐝 𝐢𝐬 𝐭𝐡𝐞 𝐫𝐞𝐚𝐥 𝐰𝐢𝐧𝐧𝐢𝐧𝐠 𝐡𝐚𝐧𝐝. Kick off re:Invent with the fastest frontier open models — GLM-4.6, OpenAI’s GPT-OSS 120B & more — all running on Cerebras Inference through AWS Marketplace (no new vendor approvals, no procurement, just pay-as-you-go). Here’s where to find us: 𝐓𝐡𝐞 𝐄𝐱𝐩𝐨 Stop by Booth 1772 in the Venetian (next to the Builders’ Showcase). 𝐓𝐡𝐞 𝐒𝐮𝐢𝐭𝐞 Want 1:1 time with our execs at the Wynn Las Vegas? Bring your use case. Leave with a plan to make it faster, cheaper, and production-ready. ➡️ https://lnkd.in/gsyShBeM 𝐓𝐡𝐞 𝐀𝐈 𝐀𝐟𝐭𝐞𝐫 We’re hosting an invite-only night for founders, builders, and AI leaders. Very limited spots. ➡️ https://lnkd.in/gmbiiMni 💥 If you’re building AI and speed is your edge, this is the week to meet us. See you in Vegas.

It is an extraordinary time to be in AI hardware. 🟧