1/ Litebeam x @reppo integration. Every call routed through the $LBM network is a real vendor interaction - a prompt sent, a response received, a payment settled onchain. Reppo turns that raw call data into a curated, independently verifiable quality layer.

GLM 5.2 is 5x cheaper than Opus 4.8 and 11x than Fable 5, yet it tops PostTrainBench. That’s exciting because lower costs make personalized intelligence economically viable. Every company and country should be able to own models trained on its own data and have sovereignty over it. The future is millions of models, each crafted around the data, values, and decisions of the people who rely on them.

So much alpha in tuning/building LLM verifiers and judges. I use them on top of my harness, and it has unlocked agentic coding workflows that are beyond anything that exists in the market today. Building verifiers and LLM judges is starting to become a skill in high demand.

Today on MCG: $REPPO | @reppo w/@rgvrmdya & @Jordan_Grollman Reppo is a decentralized evaluation and reinforcement learning layer that lets AI agents, robots, and enterprises get real-time third-party grading through a API plug in. They just crossed 500M in total volume traded on the platform, $150K in USD network fees since V2 launch in March, and 97-98% lock renewal rate. Highlights include: 02:29 - Meet the team 06:16 - @AskVenice tackles access, others tackle cost, and Reppo owns quality 09:44 - "Reppo is the Deloitte or KPMG that doesn't exist yet"...the framing that unlocked enterprise sales 10:57 - Key partnerships 12:47 - 60/40 profit share on enterprise deployments 13:00 - 50% of all off-chain revenue committed to buyback and burn 14:41 - 17 active data nets, 20-25 active enterprise conversations. Targeting 50 short-term, 100-1,000 in 6-12 months. 19:56 - "Why can't @scale_AI scale like @stripe?" 21:07 - Two lines of code + pay-per-usage → the endgame API model 38:03 - Benchmark set 42:34 - 500M+ in volume traded since V2 launch (March) (nearly $12M in USD value) 43:15 - Network fees approaching $150K 44:03 - 97-98% lock renewal rate every 48-hour epoch 50:03 - "Crypto 1000x-ing overnight isn't going to happen anymore. We're too adult of an industry now."

What does a token page look like when a team actually takes transparency seriously? I just found out. And now I can't unsee how bad everyone else's is. I say that having looked at hundreds of these. Most projects hand you a pie chart, a vesting cliff date, and call it transparent. @reppo gives you 14 allocation categories, 100% of supply accounted for at genesis, and an interactive unlock timeline you can scrub *day by day* from TGE all the way through Nov 2029. That's not a marketing page. That's the actual work. The thing that caught me... ACF unlocks on FDV milestones, not the calendar. The team gets paid in USDC, not REPPO. So they're not sitting on a bag waiting to unload at a scheduled unlock (a mechanic I've seen quietly buried in half the projects I've reviewed). The unlock only triggers when they ship. Thats the beauty of the Automated Capital Formation built into the @virtuals_io launchpad...How many projects can honestly say that? Here's what the page actually shows you: Which part of this caught your attention: the simulation builds, the AI models, or something else? 135m tokens locked in veREPPO (13.5% of supply), verifiable live at reppostats.com ➜ 3.62m permanently burned with dated receipts and tx links ➜ Protocol Safeguard Max-Lock: 8.9% of team allocation non-transferable for 2 years, enforced on-chain ➜ Burned vs locked called out separately with zero double counting (Most projects would quietly merge those last two numbers and call it "removed from circulation.") That distinction matters more than it seems... If this level of disclosure is table stakes somewhere, nobody told CT... This is what on-chain transparency looks like when a team actually builds for it. Bookmark reppo.xyz/token just to calibrate what a proper tokenomics page feels like. What's the most transparent token page you've seen outside of this?

Introducing Real-time RL. In the real world, time isn't free. The environment keeps "moving" even when you're computing your next action. We show how RL agents can learn to adaptively think in real-time games. 1/🧵

Arena reached a $100M annual revenue run rate just 8 months after launching our evaluation product. We started as a research project at UC Berkeley with a simple mission: measure AI progress through real-world use. As AI shifts from chatbots to agents taking on longer, higher-stakes work, the problem matters more than ever. Today, Arena measures real-world AI utility with a community of tens of millions. With Agent Arena, we’re evaluating long-running agents on complex, real-world tasks - how they use tools, adapt to feedback, recover from errors, and accomplish goals set by humans. We are excited to keep deepening our work in agentic evaluations. Here’s @ml_angelopoulos on what this milestone means and where we go from here: