We went quiet for three weeks. A lot happened: → Chutes has a new CEO. Christian Hyland (@cjohan247), ex-Coinbase, Toku, First Wave AI → Parallax is pretraining an 8B MoE in public. Live register, every number read from the fleet as it trains: → Jon's inference preview: 15.5k output tok/s on a single RTX 5090. Unoptimized branch, read it as a preview → Jon ran a training node off two solar blankets, a Starlink dish, and a DGX Spark in a field. Under 200 watts → New confidential VM shipped. Rebuild it yourself on any Linux box and check our measurements. Byte-identical across four builds Next week: Exploit Summit, Montreal. Jon's speaking, we're platinum sponsors, and we're not turning up to hand out stickers. Full breakdown below 👇

Announcing one year of LLM inference metadata traces, with 6.12 billion requests. We hope this dataset can support research on real-world LLM serving workload understanding, system design and infrastructure optimization. Explore the dataset and learn more: Driven by our great graduate student William Nixon and in collab with @jon_durbin @airesearch12 @chutes_ai

We gave Hermes the Chutes Agent Toolkit and one prompt. It read the docs, ran the doctor script, and wrote a provider config I could review before it touched anything. Then the test call failed: We forgot to fund my account. Prompt, config, attestation:

The push for regulatory capture and AI cabalism is accelerating. This, in large part, is why I'm building parallax. True P2P unstoppable, uncensorable AI systems aren't just neat, they're mandatory. This little single DGX spark "datacenter" in my forge with parallax can train (with other nodes) an 80b parameter model. - two solar blankets - ecoflow battery - lapdock for kvm - starlink internet - dgx spark Total input from solar ~600w (800w max) Total power draw from spark/starlink/etc. < 200w No datacenters, no water supplies, no external power grid, no multi gigabit link. We must design AI systems that are as resilient as the humble cockroach. Why do we all appreciate BTC? Because it's a mathematically certain sovereign ledger that can't be inflated away or centrally controlled. Current AI represent the greatest concentration of wealth, power, and information (IP theft? blackmail? or simply ads and behavior control?) the world has ever seen... As BTC did to finances, we must do to AI. Chutes + parallax is a step forward towards tearing down this moat.

Dashboard is a quick work in progress, but for visibility into a run here ya go! This is around $11/b tokens, insane actually. MFU also insane. The whole thing, pretty legendary, and inference... Using a few nodes from @lium_io also!

Today the Chutes team is happy to introduce our new CEO, Christian Hyland @cjohan247. Christian has worked closely with us for the past several months, developing new approaches to how we scale and operate in a fast-moving AI market. His contributions have already reshaped how we view and work within this space. With a strong background across Coinbase, Toku, and First Wave AI, paired with an extensive skill set, he was a clear choice for CEO. Excited for what's ahead with Christian in the role.

Week 16. The rundown: → One year of Chutes production traffic is now a research paper: 6.12B requests, 35.8T input tokens, 314,970 users. Co-authored with Harvard and UChicago. Request-level metadata only, nobody's prompts are in it: → The paper's finding: the users are becoming machines. Output lengths fell from hundreds of tokens to under 100 over the year. Agents call often and read short → Our TEE stack and E2E encryption layer are open source: Security you can't read is a promise. Ours is an audit away → NVIDIA researchers made the small-model case for agents. Route, extract, format, call a tool. Small-model work at small-model prices Full article below 👇

Astroware Labs (@astrowareai) and @trishoolai are plugging Halo Guard Alpha into Chutes Chat. It's a safety classifier that catches jailbreak attempts. Astroware trains it against live attackers on Trishool (SN23), where they pay miners to break the model and feed every successful break back into training. Halo hits 87% F1 across 7 public safety benchmarks including Aegis and HarmBench, and catches attacks like narrative injection and encoding obfuscation. Another step forward in AI Security.