With less than five days of notice, Anthropic decided to cut off nearly all of our first-party capacity to all Claude 3.x models. Given the short notice, we may see some short-term Claude 3.x model availability issues as we have very quickly ramped up capacity on other inference

This 12B model runs Qwen3-style reasoning on a single GPU. So you get Qwen3-style CoT traces without the 235B overhead. Homunculus-12B distills Qwen3-235B onto a Mistral-Nemo backbone. It mimics how Qwen thinks. huggingface.co/arcee-ai/Homun…

NVIDIA just released Llama-Nemotron-Nano-VL-8B-V1, an 8B vision model that reads dense documents, charts, and video frames. It's #1 on OCRBench V2 (English), with layout and OCR fused end-to-end. huggingface.co/nvidia/Llama-3…

From "𝗪𝗲 𝗖𝗮𝗻'𝘁 𝗔𝗳𝗳𝗼𝗿𝗱 𝗛200𝘀" to "𝗛200𝘀 𝗔𝘃𝗮𝗶𝗹𝗮𝗯𝗹𝗲 𝗡𝗼𝘄" Every founder knows this moment: You need something expensive, and your bootstrap budget laughs at you. Six months ago, that was us with H200s and H100s. Our customers needed them (thanks to

Now we are reaching 500K monthly downloads 💪

super excited to announce myrunningpal ! as a runner myself often i feel the need of a coach who can guide me live based on my metrics like heart-rate, cadence, vertical oscillations etc.. i made ai do that and now i am launching it for public use ! join the waitlist now!

𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗔𝗜 𝗶𝗻 𝗜𝗻𝗱𝗶𝗮 𝗶𝘀 18 % 𝗺𝗼𝗿𝗲 𝗲𝘅𝗽𝗲𝗻𝘀𝗶𝘃𝗲 𝗯𝗲𝗳𝗼𝗿𝗲 𝘆𝗼𝘂 𝗲𝘃𝗲𝗻 𝘄𝗿𝗶𝘁𝗲 𝗮 𝗹𝗶𝗻𝗲 𝗼𝗳 𝗰𝗼𝗱𝗲. The culprit isn’t talent or power costs. It’s GST on GPUs. While global rivals pay near-zero, Indian builders get taxed for innovating

an attempt at cracking memory for multimodal models context: recently i have been reading about memory layer for ai agents, and i wrote this blog fullstackagents.substack.com/p/forget-me-no… this got me thinking about memory layer for multimodal models and then i recalled project astra by Google.

Interesting insight, based on the knowledge available I assumed sglang is faster than vllm

Can’t agree more, Indian founders need a global voice, has to aspire global from day 1

🎉 Finally Cursor has notebook support!

When you stop optimizing for an end goal, you also stop getting anxious about whether you’re doing the right thing. When you don’t care about the destination, the specific path you’re on becomes as good as any other path you might chose.

Vishnu - Jarvislabs.ai Ashwini Vaishnaw Sir, this is a very valid concern. We need to remove all taxes on the bleeding edge GPUs to keep Indian deep tech competitive.

Best time to be curious 🧐 Thanks to tools like ChatGPT, Claude I am able to ask questions like a kid

Do you know what is the number 1 reason for a startups to die? When the founder gives up. Founders probably get a million reasons to do it. I learnt this from a Ycombinator video long back, when things go bad I alway remember this and keep pushing.

Connecting dots backward, I heard about it for the first time when I read the book on Steve Jobs. Post that, when I have to do something in life, some of them could be boring, I always remember the "Connecting dots backward" statement and try to move on. A few months backs, I

Moonshot AI just opensourced Kimi-Dev-72B! Achieved 60.4% on SWE-bench Verified, which is SotA among opensource models

Meet Jan-nano, a 4B model that outscores DeepSeek-v3-671B using MCP. It's built on Qwen3-4B with DAPO fine-tuning, it handles: - real-time web search - deep research Model + GGUF: huggingface.co/collections/Me… To run it locally: - Install Jan Beta: jan.ai/docs/desktop/b… - Download