Together with NVIDIA we post-trained Nemotron 3.5 Super VL on thousands of tasks Reflecting enterprise workflows in sales, operations, marketing, finance We improved the AutomationBench-AA score by ~70% on the public split

one of my widely held beliefs is that open models are way more capable than ppl (and benches) realize this project is one of many that reinforced my beliefs. an incredible team + a few tokens are all you need : )

This has been insanely fun to work on for the past couple weeks! This rewrite tested the limits of all our products: thousands of subagents via Prime Agent, billions of tokens via Prime Inference, and hundreds of sandboxes via Prime Sandboxes! It's incredibly rewarding to see our full stack come together, and Prime Agent is now faster than ever :)

Over two weeks, Prime Agent orchestrated a swarm of over 2,000 agents to rewrite itself in Rust. With 10,000+ sandboxes, 200B+ GLM-5.3 tokens, and 16,000 agent-to-agent messages, this was our largest test yet of Prime Agent’s multi-agent capabilities and our infrastructure powering these agent swarms. Prime Agent now reaches usable input ~13× faster and uses 83% less startup memory than before.

i'm giving free Prime Sandboxes credits for qualifying startups through Series A the same sandboxes our researchers use for RL. high concurrency needs encouraged. if you're training or running agents at scale, DM me what you're building and why existing solutions don't fit.

Inference scaling works, but every agent eventually faces the same problem: its context window fills up. In this essay, we argue the only scalable fix leads straight to swarms.

Introducing Prime Inference: We've served trillions of tokens for RL and dedicated customer deployments To own your intelligence, you need to own your inference Unpacking our inference stack

Using Prime Intellect, Extropic post-trained Qwen3.6-35B-A3B for thermodynamic ML research, nearly tripling its eval results on held-out tasks in ~100 GRPO steps. They built a custom RL environment with verifiers and trained on Hosted Training, Prime Sandboxes, and Prime Inference. This meant Extropic didn't have to manage multi-node GPU infrastructure, so the team could focus on research.

Introducing Prime Sandboxes: MicroVM sandboxes purpose-built for RL training. Model training requires running tens of thousands of concurrent sandboxes, leading to complex and costly configuration. We built Prime Sandboxes for our own team. Today we're releasing them publicly.

The Goodfire team used Prime Intellect to train activation probes to detect reward hacking. With them, they are able to catch reward hacking in various models. Their probes are performing similarly or better than frontier LLM-as-judge setups, while being more efficient.

we are super bullish on sparse attention + HiSparse and have been working with the vLLM team on it sparse attention reduces the pressure on memory bandwidth by only selecting k for the attention, but it doesn't reduce KV cache memory storage , in high-throughput wide-EP deployment you want to maximize the batch size of decode to use compute as much as possible, but at long sequence decode you quickly run out of VRAM and can't hold enough parallel requests to saturate the compute HiSparse fixes this by offloading the active KV cache to CPU. It keeps an LRU cache on GPU, and since many of the same K tokens are reused every decode you barely notice the offloading, this allows a massive decrease in memory usage and an increase in concurrency This is super important for RL where throughput is key and we want to be as much as possible in a compute-bound regime tldr: lower memory usage, more concurrency, higher inference throughput, faster RL

Sparse MLA only attends to the top-K tokens, so the rest of the KV need not live on the GPU. Hybrid HiSparse in vLLM builds on that, and a request keeps decoding after its KV stops fitting in HBM. It keeps KV on the GPU while there is room. Under pressure a request releases its coldest pages to host memory, keeps a small hot buffer of what the indexer asks for, and keeps decoding instead of being preempted. 📊 Demonstrated on GLM 5.3, one 8× H200 node, full 1M context. Same host memory, configured concurrency 32: KV offloading kept 5-6 requests running. Hybrid HiSparse kept 19-25. 🔹 Hot pages are ordinary KV blocks from the same pool (Hybrid Memory Allocator) 🔹 One fused kernel resolves resident, hot and missing rows, CUDA-graph capturable 🔹 Prefix caching, OffloadingConnector, P/D imports and MTP keep working Built by @RedHat_AI and @PrimeIntellect with the vLLM community. Planned for v0.30; pinned commit, flags and calculator are in the post👇 🔗