For agent builders, the important part is simpler than the protocol: let software pay only for the service it needs, then get usable access immediately—without an account-creation or API-key workflow. L402 makes that possible on Lightning. ⚡

Venice is going vertical. We're standing up B300 GPU clusters in our own data centers and we need a Venetian who lives at the intersection of CUDA kernels, quantization math, and production inference at scale. If you've squeezed @vllm_project or @sgl_project past their limits, profiled with Nsight, shipped tensor-parallel multi-node configs, and want to do it all in service of privacy-first AI infrastructure for millions of users, this is your call. Come build the Port City of AI with us.

As we mentioned, success rates stayed close, but speed and token efficiency split between harnesses: - Success rate: 22/28 for Kimi Code, 21/28 for Hermes, 20/28 for Claude Code - Median time per task: 179s in Hermes, 297s in Kimi Code, 348s in Claude Code So the fastest harness (Hermes) and the most token-efficient one (Kimi Code) were different.

The median task consumed nearly 6x more tokens in Claude Code than in Kimi Code: - 61k in Kimi Code - 67k in Hermes - 340k in Claude Code At K3's $3/M input rate (input tokens make up roughly 95% of agentic workloads), the average cost per task was: - $0.22 in Kimi Code - $0.28 in Hermes - $2.00 in Claude Code

Kimi K3 can now be run locally! ✨ The 1-bit model retains ~78.9% accuracy after we shrunk it from 1.56TB to 594GB (-62% size). Run on a Mac Studio + 128GB RAM device. Kimi K3 is the strongest open model to date. Guide: GGUF:

i posted this a couple weeks ago about needing a second dgx spark, and nvidia just shipped me one. the unit and the connectx cable, on a truck to bangkok within a week of me saying it out loud here on x. sit with that for a second anon. the biggest company on earth moved that fast for one guy in a room in bangkok. no stanford lab connection behind me, no committee, no comms team drafting my posts, just a builder putting real numbers in the open where anyone can tear them apart. the polished accounts with the pedigrees have been saying "democratize ai" for years, nvidia shipped the box to the guy actually doing it. giants aren't supposed to move like this, and that's exactly why the open side feels different right now. nvidia is carrying real weight for the local ai side, backing the people who own their compute instead of renting cognition from an api, and doing it quietly, on the ground, for whoever's actually doing the work. i think that's the whole reason this movement has legs man. here's what happens next. two dgx sparks, one connectx cable, 256 gigs of unified memory, tensor parallelism across both. the models that never fit on one box, glm 5.2, nemotron, deepseek v4 flash at a million tokens, they gonna clank now. i'm going to learn a stupid amount from this setup man, and share all of it, same as always. biggest shoutout from deepest depth of my heart to @Coolmark482 for making all this happen and for seeing the local ai community as worth betting on. thank you.

There sure has been a lot of *talk* about Bitcoin development these last 18 months. Meanwhile, there's been an awful lot of actual development work done. Examples of what Bitcoin Core has been working on throughout this all: - a node that syncs in nearly half the time - a path to 5x faster (SwiftSync) - a new GUI to make running that node easier, maybe even on mobile - a redesigned mempool with better fee bumping and fee estimates - a mining interface so individual miners can build their own block templates (Stratum v2) - a private broadcast so your transactions can't be traced to your IP - a reproducible map of the entire internet for node eclipse attack protection - a better way to relay time-sensitive Lightning and Ark transactions for confirmation - a multisig that's cheaper, more private, and looks like any other payment on-chain (MuSig2) - a private payment address with zero linkable payments (silent payments) - a threshold (e.g. 3-of-5) signature scheme that looks like a single signature (FROST) - a proposal to fix known protocol bugs and current attack vectors (BIP54) - a slew of post-quantum research to be ready for future attack vectors - a first-ever independent security audit for more eyes on the code - a new tool that attacks a live node the way a real adversary would (Fuzzamoto) - a fix that killed an entire class of disk-filling attacks - a replacement for a risky external library - a consensus engine the whole ecosystem can build on (Bitcoin Kernel) Over a hundred contributors and more than a dozen funding organizations around the world. Not without controversy. Not without bugs. But undeniable progress. August will come and go. This work will continue.

Introducing NostrWolfe Bridge 🐺 Live, searchable service listings in Buzz—where agents already work. The bridge stays walletless. When agents are ready to buy, Lightning Enable MCP can pay an L402 endpoint within a budget you control. Open source:

Nous Portal + Buzz = 4 Agents on a team, powered by FREE models in the Nous Portal, debating which is using the best model 😂 If you're managing multiple agents, @NousResearch + @blocks Buzz has to be one of the best options out there right now!

Hermes Agent data for the past 30 days usage on OpenRouter shows DeepSeek V4 Flash still at the top followed by Hy3 from @TencentHunyuan other notable models like Step 3.7 Flash and MiniMax M3 are on the board... which model is your fav to use with hermes agent?

Hidden Hermes gem: /moa Most of the time you ask your agent a hard question and get one answer. What if that answer is wrong… and you have no way to know? Hermes has a feature called /moa (Mixture of Agents) With /moa , Hermes sends the same prompt to 2+ different models in parallel. Each one thinks independently. Then a third model (the aggregator) reads all their answers and synthesizes the best one. The best part? It still works inside the normal Hermes agent loop. Tools Sessions Follow-ups Interrupts /goal And when that one task is finished, Hermes switches back to the model you were already using. Real example: I hit a nasty TypeScript error. One model said recursive loop. Another said constraint issue. The aggregator compared both perspectives, found the actual root cause, and gave me three working fixes. Set it up with: hermes moa configure Pick your reference models and aggregator. Then use: /moa your difficult question Or select your MoA preset like a normal model in Desktop, CLI, TUI, or the Dashboard. Yes, it uses more model calls. Yes, it can take longer. But for architecture, debugging, research, or any decision where being wrong is expensive… I would rather have a panel than trust model luck. The HermesBench result is wild too: MoA: 0.8202 Opus 4.8 alone: 0.7607 GPT-5.5 alone: 0.7412 Not one model to rule them all. A team of models working together.

New York spent twenty years getting oil out of its power plants, from 17 million barrels a winter in the mid-2000s down to almost nothing. That progress just reversed. Last winter: 4 million barrels, the most since 2008–09. January alone was the biggest oil-burning month in nineteen years. And the response wasn't a crash program to fix the cause — it was a new grid-operator rule that pays plants to keep oil on site, with penalties if the tanks are empty. 80% of downstate's fossil fleet signed up. Sign-ups for the winter after next close this Friday. Germany's Energiewiede brought coal-burning back; New York's CLCPA is bringing back oil.

This is the model @ambosstech Payments is built on: bitcoin, USDT, and USDC over Lightning, settled straight to infrastructure you control. 0.5% flat. Global. The gateway can't hold your money hostage because it never holds your money.

@BitcoinWhitcomb @Bitlatethenever @BigpictureBTC @matbalez @shoneanstey @freddienew @KyleHutch_ @davelund_ @LQWDTech @LQWDClaw @spiralbtc @CashApp @OpenAI @z @LangChain Built an HTLC Dashboard to help monitor and improve routing performance....🔥 Hermes Agent FTW ✅