Swapnil Bhatkar

I sent a DM to @kelseyhightower from Google about my concerns related to public speaking. He quickly responded and sent me a Meet link. We spoke for about 40 minutes around 2.30 am ET. He gave me some of the best advice I could’ve ever received. A thread 🧵

Built a Rust engine from scratch for HPC Slurm scheduler with deterministic artifact generation, policy validation & solid test coverage. ~5K LOC in a single session. GPT-5.3-Codex model is impressive and fluent in Rust. @OpenAIDevs really cooked with this one @gdb @thsottiaux

@dabit3 @cognition @DevinAI @windsurf Congrats Nader. We needed you back here 🎉

Life update - I'm joining @Cognition! One goal: end-to-end software agents that ship production code. Cognition covers the entire development lifecycle. I'll be working with @DevinAI and @windsurf, focused on education, growth, awareness, and developer experience.

🚨The White House just launched the Genesis Mission — a Manhattan Project for AI The Department of Energy will build a national AI platform on top of U.S. supercomputers and federal science data, train scientific foundation models, and run AI agents + robotic labs to automate experiments in biotech, critical materials, nuclear fission/fusion, space, quantum, and semiconductors. Let’s unpack what this order actually builds, and how it could rewire the AI, energy, and science landscape over the next decade:

Announcing the completely reimagined vLLM TPU! In collaboration with @Google, we've launched a new high-performance TPU backend unifying @PyTorch and JAX under a single lowering path for amazing performance and flexibility. 🚀 What's New? - JAX + Pytorch: Run PyTorch models on TPUs with no code changes, now with native JAX support. - Up to 5x Performance: Achieve nearly 2x-5x higher throughput compared to the first TPU prototype. - Ragged Paged Attention v3: A more flexible and performant attention kernel for TPUs. - SPMD Native: We've shifted to Single Program, Multi-Data (SPMD) as the default, a compiler-centric model native to TPUs for optimal execution. Dive deep into the new architecture and see the performance benchmarks in our latest blog post! blog.vllm.ai/2025/10/16/vll… #vLLM #TPU #JAX #PyTorch #AI #OpenSource

Federal workers deserve access to the most capable AI tools to better serve the American people. Today, we’re removing cost barriers to Claude for all three branches of the U.S. government.

Can confirm. Claude Code with Max plan ($200) is heavily nerfed. It’s unable to do the same thing which it did 2 months ago with exact same CLAUDE.md @alexalbert__ @_catwu @dmwlff

@thorstenball @thorstenball Loved your blog on building agents with Go! Genuine question: Why aren’t more teams using Go for CLI agents?

Time to update your Amp CLI

@nikunjhanda Will web search tools and deep research models be available in Azure anytime soon?

So excited to be launching 4 features in the OpenAI API today: 1. Deep research with support for web_search and remote mcp (for your data that's not publicly accessible on the internet) 2. Web search in o3, o4-mini, and o3-pro — at a new reduced price of just $10/1K queries 3. Webhooks support in responses, batch, fine-tuning jobs, and evals runs 4. Logprobs now supported in Responses API (sorry it took so long)

Here's a full blog post on Code with Claude developer conference in SF last week Claude Code is now GA with full IDE integration and SDK availability for headless mode. Shoutout to @_catwu @bcherny @ErikSchluntz Full writeup: swapnilbhatkar.cloud/blog/code-with… #CodeWithClaude #Anthropic

Attended @AnthropicAI first developer conference #CodeWithClaude in SF today. What a time to be alive! The best agentic coding model out there. Got first class access to research demos and 1:1 interaction with creators of Claude Code #Claude4

@mikeyk Is this available on Bedrock and Vertex?

Just shipped web search to our API! We already use this in Claude Code to dynamically fetch documentation or other info, now you can build it into your apps too.

@thorstenball This is refreshing. Great job explaining this and making it look so simple

Read it here: ampcode.com/how-to-build-a…

Finally did it. I wrote down how to build a code-editing agent. In 315 lines of code. And yes, it works. Very well. There is no moat.

@awilkinson @awilkinson your wish will be answered soon. Screenshot from iOS app store today 👇🏻

I would pay $500 a month to not hit Claude's limits. Please god.

Introducing our first set of Llama 4 models! We’ve been hard at work doing a complete re-design of the Llama series. I’m so excited to share it with the world today and mark another major milestone for the Llama herd as we release the *first* open source models in the Llama 4 collection 🦙. Here are some highlights: 📌 The Llama series have been re-designed to use state of the art mixture-of-experts (MoE) architecture and natively trained with multimodality. We’re dropping Llama 4 Scout & Llama 4 Maverick, and previewing Llama 4 Behemoth. 📌 Llama 4 Scout is highest performing small model with 17B activated parameters with 16 experts. It’s crazy fast, natively multimodal, and very smart. It achieves an industry leading 10M+ token context window and can also run on a single GPU! 📌 Llama 4 Maverick is the best multimodal model in its class, beating GPT-4o and Gemini 2.0 Flash across a broad range of widely reported benchmarks, while achieving comparable results to the new DeepSeek v3 on reasoning and coding – at less than half the active parameters. It offers a best-in-class performance to cost ratio with an experimental chat version scoring ELO of 1417 on LMArena. It can also run on a single host! 📌 Previewing Llama 4 Behemoth, our most powerful model yet and among the world’s smartest LLMs. Llama 4 Behemoth outperforms GPT4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro on several STEM benchmarks. Llama 4 Behemoth is still training, and we’re excited to share more details about it even while it’s still in flight. A big thanks to all of our launch partners (full list in blog) for helping us bring Llama 4 to developers everywhere including @huggingface, @togethercompute, @SnowflakeDB, @ollama, @databricks and many others👏 This is just the start, we have more models coming and the team is really cooking – look out for Llama 4 Reasoning 😉 A few weeks ago, we celebrated Llama being downloaded over 1 billion times. Llama 4 demonstrates our long-term commitment to open source AI, the entire open source AI community, and our unwavering belief that open systems will produce the best small, mid-size and soon frontier models. Llama would be nothing without the global open source AI community & we are so ready to begin this next chapter with you. 🦙 Read more about the release here: llama.com, and try it in our products today.

@alexalbert__ @AnthropicAI That’s cool. Any word on when the Go SDK’s going GA?

Good news for @AnthropicAI devs: The Anthropic Java SDK is now generally available!

🚀 NVIDIA GTC 2025 blog I spent an incredible week in San Jose attending keynote, hands-on sessions, and workshops covering everything from multimodal and reasoning AI models to CUDA magic and new GPUs 👉 Read the full recap: swapnilbhatkar.cloud/blog/nvidia-gt… #gtc2025 #NVIDIA #vllm #aiops

MCP 🤝 OpenAI Agents SDK You can now connect your Model Context Protocol servers to Agents: openai.github.io/openai-agents-… We’re also working on MCP support for the OpenAI API and ChatGPT desktop app—we’ll share some more news in the coming months.

@AravSrinivas Well earned 🙌🏻

Honored to have Jensen say great words about Perplexity! We’re going to be doing a lot together with NVIDIA on inference on Blackwell with their new Dynamo library!