Lesheng Jin

Databricks ranks #1 on NVIDIA’s SOL-ExecBench kernel leaderboard, in the L1 single operation track, powered by KDA (Kernel Design Agents) 🎉 What’s crazy is: we 100% leveraged AI agents to beat the competition. This is a sneak peek at recursive self-improvement. The core frameworks we used were KDA, Humanize, and Omnigent: Claude writes code, Codex reviews. Together, they enabled agents to run autonomously for as long as possible. The key is setting up the right framework to let the agents cook. This work was driven by @leshenj15 at Databricks, in collaboration with NVIDIA and MIT HAN Lab’s @LigengZhu and @DongyunZou03 . Databricks AI is like a neolab. Join us if you’re cracked!

You may have heard that GLM-5.2 at 328 token/s is cool, How about 392? Databricks is now #1 in inference speed for GLM-5.2 on Artificial Analysis. It's a great model, and we did a lot of optimizations.

An OpenAI friend told me he burns 300M GPT-5.5 tokens/day. The top one in his team burns billions of tokens/day. Codex coding for them every night. Databricks also gives engineers unlimited tokens. We're looking for cracked inference engineers to join us at Databricks AI to produce trillions of tokens, insanely fast. DM me if you have: - Contributed to open-source ML systems like SGLang/vLLM/PyTorch - Experience serving LLMs at large scale Databricks AI runs like a startup. Lots of exciting things to build!

@leshenj15 Thank you Lesheng, for all your contributions to Hyperbolic! The AI inference product would not be where it is today without you. Hope our paths cross again in the future. 🫶 All the best.

I have decided to step down as CTO at Hyperbolic. Leaving a company you co-founded and poured your heart into is not easy. So many moments still feel vivid: launching our AI inference product for open-source models and seeing tens of thousands of developers sign up in a week; the week we were hit by a massive DDoS attack and the entire engineering team fought around the clock until we won; the day we launched the GPU platform and watched ARR take off. There were also hard moments. That’s the nature of building a startup. I’m grateful for all of it. What I’m most grateful for is the team. Thank you for your trust. Most startups never build something people want. I believe we did. You should be proud of yourselves. I will look forward to seeing your success. What’s next for me? I’m still figuring it out. I believe this is the most extraordinary moment in human history. We’re standing at the edge of the Singularity. AI will reshape everything, and I still feel the same excitement I felt when I first fell in love with AI. Time to start over. Time to climb another mountain. Thank you to everyone who has been part of the journey, — Yuchen

@junrushao 严肃程序员

@Yuchenj_UW new interview: debug the code generated by llama

Meta finally abandoning LeetCode for AI-assisted coding interviews. Biggest big-tech advancement of 2026.

@AlexanderTw33ts @ChatGPTapp banger

@ChatGPTapp Here I fixed it

@Yuchenj_UW @_thomasip lol

@_thomasip Wrong. I have a PhD lol.

Topping Today’s News with this was not on my bingo card. (I beat Sundar Pichai!) To be clear, PhDs have made essential contributions to tech innovation, no doubt. But the degree itself is not required to drive innovation. Passion, agency, taste, resilience >> degree.

@Yuchenj_UW Meta fired a bunch of talented employees to cover up the fact that too many talented people had left.

Meta laid off 600 people from its Superintelligence Lab today. Many FAIR researchers, including FAIR Research Scientist Director Yuandong Tian, were affected. I think Yann Lecun will leave soon. Maybe I should raise $2B and start a new frontier lab with these folks.

@simon_zhai @Yuchenj_UW lol

@Yuchenj_UW only children make selections, men want them all

closed-source AI vs. open-source AI which way western man?

We just launched On-Demand GPU Cloud at Hyperbolic! - H100s at $1.49/hr - reliable nodes with persistent storage - spin up a cluster of GPU nodes with Infiniband in just a few clicks Tell me what you're building. I'm giving out free credits to cool projects.

Looking back, joining hyperbolic is one of the best decisions I've made! I'm proud to be a part of this talented team.

🚀 Excited to share that we raised a $12M Series A! At Hyperbolic, our mission is to build an open AI platform. By "open", we mean: > Open GPU marketplace: Think of it as GPU Airbnb—anyone can contribute their GPUs. We’ve started providing H100 SXM at $0.99/hr, the lowest price you can find anywhere. More people from "GPU poor" to "GPU rich" means more AI innovation. > Open-Source Models: We believe open-source is the future of AI. We’re committed to hosting and supporting these models, and since launching our inference service 5 months ago, more than 40,000 developers have used us. > Open community: Transparency and collaboration are at the heart of what we do. By building in public, we hope to create maximum value and the best products for our users. Let us hear your feedback! Now we continue to accelerate! 🌪️

@syeddhasnainn @Yuchenj_UW @hyperbolic_labs @Alibaba_Qwen @openrouter Thank you for the feedback! I tried on my side, and the result seems good. This result might come from other platform. Did you choose Hyperbolic as the only provider on OpenRouter? Let me know if you meet any issues.

@Yuchenj_UW @hyperbolic_labs @Alibaba_Qwen @openrouter yeah right

Happy Thankshipping! We @hyperbolic_labs now serve QwQ-32B-Preview released by @Alibaba_Qwen today in BF16! 🚤 > A reasoning model that is competitive with OpenAI o1-mini and o1-preview. > Integration with @openrouter coming soon. > It can answer the Strawberry question correctly every time (we auto-applied the system prompt for the API). Throw your hardest problems to it and compare it with o1 and Deepseek-R1. It's incredible to see China taking the lead in open-source models that can reason, especially given the challenges of working with fewer and worse GPUs than US labs. Kudos to @JustinLin610, @huybery, and the Qwen team for what they are doing for open-source AI.

Launching @hyperbolic_labs "hyperbolic-gradio" package, and our first HuggingFace spaces app similar to Claude Artifacts. Here's how I built a tic-tac-toe game using the app, with just a prompt. Shoutout to @_akhaliq and the @huggingface team for setting up the hyperbolic-gradio package. It makes deploying a @Gradio app with Hyperbolic Inference API incredibly simple. This app leverages the package to call Hyperbolic Qwen2.5-Coder-32B-Instruct bf16 API to write code.

@AnalogPvt @YayaSoumah @hyperbolic_labs @JustinLin610 @huybery Frontend stuff. It's 0.2 now. Thanks for the info!

@YayaSoumah @hyperbolic_labs @JustinLin610 @huybery Why price is higher on hyperbolic labs

With Hyperbolic providing 128k output tokens for Qwen2.5 32b, I can see a tremendous value for making large updates all at once. Can't wait to test this model deeply. I'm very curious about the prompt adherence. @hyperbolic_labs . Congrats to the team @JustinLin610 @huybery @openrouter

Hyperbolic’s inference service has hit over 1 BILLION LLM tokens daily! This achievement highlights the growing demand for our AI services and the trust our community members place in our platform. With 22 models (and counting), we’re dedicated to offering top-tier AI models at a fraction of the cost. 💜 Build with us: app.hyperbolic.xyz/models?utm_sou…

Introducing the Hyperbolic Ambassador Program At Hyperbolic, we believe in a future where AI is open source, decentralized, and accessible to all. This program engages those who are committed to decentralizing AI while supporting career growth. Earn up to $1,000 in platform credits and $2,400 in meetup credits. Whether you're a startup founder, university researcher, developer, or simply enthusiastic about AI, we want to hear from you. ➡️ Apply by October 9th, 2024 share.hsforms.com/1Yk1jYvSrSQu9y…