Periodic is hiring SWE's for physics simulations, scientific data, and lab robotics These are hard problems that codex/cc can't one-shot, because they require interfacing with the real world and collaborating with scientists Science experience is a + but not required! Apply ⬇️

🚀New blog: Accelerating Long-Context and Agentic Inference with NVFP4 KV Cache 4-bit KV cache is here! NVFP4 KV in SGLang packs ~1.78× more context into GPU memory and speeds up long-context decode by up to 78%. Together with @Alibaba_Qwen and @nvidia, we brought NVFP4 KV cache to Blackwell: - NVFP4 stores KV in just ~56% of FP8's footprint per token -️ Decode throughput jumps +37% / +58% / +78% at 32K / 160K / 1M context - Near-lossless accuracy: matches FP8 on GPQA-Diamond & AIME 2025 (Qwen3.5-397B-A17B) - Higher cache hit rate keeps AgentX throughput scaling where FP8 drops off The whole recipe combines NVFP4 two-level scaling, in-kernel dequantization on decode, and paged KV cache, and you can enable it in SGLang with a single flag --kv-cache-dtype nvfp4

KEYNOTE: Inference economics increasingly shape what’s possible in AI. At Fellows Forum 2026, @BanghuaZ, co-founder & CTO of @RadixArk, will give our Day 1 keynote on the economics of inference and what it takes to run models efficiently at scale. RSVP:

.@periodiclabs is an insanely talented team, tackling some of the hardest problems in science, and delivering awe-inspiring progress at the frontier of AI and physics at remarkable speed. Read about their work below!

Neon’s scientific traces are long and vary widely in length. Training on them shouldn’t mean GPUs chewing through padding or waiting for each other. Here’s how we pack sequences, balance the work, and distribute long contexts across GPUs at @periodiclabs 👇

Another Neon infra tidbit: @periodiclabs accelerated checkpoint conversion ~30x down to just 1 minute, helping us deploy faster to get rapid feedback from our labs. Our models train in Megatron, but SGLang inference consumes weights in HuggingFace format. Traditionally, conversion ran serially: 1. rebuild the entire checkpoint from individual weight tensors 2. convert the checkpoint 3. reshard it to be inference-ready This could take 30 min for a trillion parameter model! Noticing that conversion shouldn't require materializing the full checkpoint, @hsu_byron introduced Fast Resharding: 1. parallelize conversion across Ray actors 2. convert expert chunks directly instead of assembling the full expert tensor in memory We upstreamed Fast Resharding to Miles in PR #1371#. Now even you can deploy trained models to production in a matter of minutes.

in the early days of @AnthropicAI so much had to be done from scratch it’s a miracle the company exists at all however, now, so much is off the shelf entirely new rates of frontier progress are now within reach for new labs all you need is the right mindset

You saw the AI & science. Let's talk about the RL infra it took to build @periodiclabs Neon. To minimize training-inference mismatch in RL, SGLang captures inference's MoE routing decisions for each rollout and we "replay" them while training. In agentic (multiturn tool-use) settings, SGLang exports these router decisions in response to each decoding request, i.e. after each conversation turn. So when *any* data-parallel rank finishes a conversation turn, *all* other ranks must wait until routing data finishes exporting. This slowdown is exacerbated because we export routing decisions from the *entire* conversation rather than just the most recent turn! When @hsu_byron @vwxyzjn discovered this in our Kimi K2 RL setup, they introduced Delta Router Replay: cache previous turns' router decisions on the training client, so you can export only the delta (most recent turn's router decisions) upon each decoding request. Delta Router Replay significantly speeds up our long-context agentic RL runs, and @hsu_byron upstreamed it to SGLang (#24851#) a few months ago.

Nearly half a year of silence. We spent it studying one problem: how far RL can scale. MiMo-V2.6 is in the middle of its RL run right now. Three things we scaled: compute (~2B tokens per step, 1568 prompts × 16 rollouts, fully async), environments and harnesses (multi-task agentic RL, mixed across multiple harnesses in one run), and grader compute (agentic in-group credit assignment, with test-case and rubric-based rewards). We'll open-source the details piece by piece over the coming weeks. Streaming the run:

I am reading the exact opposite way. This effort is very cool, and isn't it kind of amazing to collect your own data, run a custom midtrain, and then running your own RL to end up at the same performance as Astra max for 1/2 the inference price? Imagine what can then be next.

We check OOD generalization for @periodiclabs Neon and report optimal cost-performance here, too. This indicates useful generalization for our model. But our experimental data also opens up an interesting ML research program. For example, RL tasks to directly predict experimental outcomes given all prior experimental data known up until that date. This is loosely analogous to the often-proposed ML experiment of training AI on all literature before 1905 and then evaluating whether it can derive the theory of relativity. An AI that already knows the answer through pre-training can cheat on RL tasks like this (i.e., it can skip reasoning and simply output a memorized experimental result). A unique, complete, dated system of record makes this work possible and fruitful.

@zijie_y is absolute GOAT of frontier-level training. And he's looking for a new colleague to help him even further increase our science/gpu-hour. Job posting to appear soon, for now you can DM me or him

Periodic Labs is doing extremely cool work applying frontier AI research to materials science. Here they post-train a model for a X-ray diffraction analysis, beating GPT-6 Astra and boosting performance of Kimi K2.6 by 20x! What's worth noting here is Periodic is doing RL directly with real-world experimental data, which comes with many unique challenges over coding/knowledge-work environments.