🚨X1vault.com Permanent storage is live on #X1 chain! Most of what you put online stays up only as long as someone keeps paying for it. If you stop paying, change hosts or lose an account, it’s gone. X1vault.com fixes it! Whatever you upload to X1vault.com, whether it's a text file, image, document, code, audio, video, or a web page, will be forever stored and it'll live as long as the #X1 chain exists. No one can remove it, or lose it! These stored files are your assets, too. You can list them for sale or buy what others have inscribed. Note: The inscription numbers claimed right now are the lowest that will ever exist. More importantly, every upload contributes a small fee into a shared rewards pool, which is paid out monthly to the top 3 uploaders and 2 randomly selected accounts. Reward pool already holds💯 neptune-privacy:native and we’ve barely just started. Go ahead and try it, every upload is a ticket to win! It's easy to start: ➡️ Go to x1vault.com → connect wallet → store a file. Then you own something that can’t be taken away from you and it'll live on #X1 for as long as the chain exists. #X1 #XNT Coming soon to @solana!

12 hours in and 305 $XNT already waiting in the x1vault.com reward pool. Pool’s still filling. Make just one permanent upload on #X1 and you’re in the draw. 5 wallets will split the price, 2 of those are picked randomly by a block hash nobody can rig.

Whoa, we launched x1vault.com that allows you to permanently store files on X1 chain... and in just a couple of hours, people uploaded 111 files. These will be forever stored on the #X1 chain. Also, there is over 204 $XNT now up for grabs in the Reward Pool, where the top 3 uploaders are guaranteed to split 78% of the pool, while 2 random uploaders will share the remaining 22%. So, every uploader has a shot at winning. I am impressed guys... I see images, videos, audio, html files, playable html games, json files, encrypted and compressed files... Many of these are listed for sale. Btw. we just had our first sale on the marketplace! Happy to see the traction.

I've tested @coldniko's latest release of 125B MoE coder (Strata's Qwen3.8-Flash-Next Coder - IQ1_M) on a single RTX 4090 GPU with 64GB of RAM and it writes code twice as fast as a QWEN 3.8 27B dense baseline, but unfortunately, it's not as good at coding, it has only half the autonomy of Qwen3.8-27B (UD-Q4_K_S) and you're better off to stick with Qwen3.8-27B. On speed alone Strata's Qwen3.8-Flash-Next Coder wins: 87.9 vs 43.3 tokens/s. This is thanks to activating only 10 experts per token and speculative decoding, the 125B model finished most simple calls in ~4.8s vs 10.7s for Qwen3.8-27B. In algorithmic tasks with reasoning enabled, Strata went 3/3 green in ~1.2 minutes vs the 27B model's ~3.3 minutes. So also good, but on the autonomy ceiling is just couldn't compete: 8/21 green (38%) vs 33/40 (83%): In other words, on simple tasks it was dependable. 100% on one-line reasoning, 3/3 on a simple Roman numerals test case, 2/3 on CLI data processing. But on Interactive DOM apps it was a total collapse. It went 0/9 across a couple of test for apps like: Todo list, Memory card game, Sudoku and Tetris. Now, why It fails? I think the reason why it wasn't as good is the aggressive 1.75-bit quantization. While that doesn’t cause complete gibberish and code looks good on a first look, if you look deeper, it causes subtle local slips: - Writing solve-btn instead of required ID solve - Classic JS closure traps in loops - Leaving event handler stubs empty (onKey(e) {}) On self-tests 4 of the 8 product bugs passed the model’s own generated unit tests. It marked them "Goal Met", but the independent referee caught the dead UI and app not working. So my verdict is that engineering-wise, running a 125B MoE at ~88 tok/s on 22 GB VRAM + 28 GB RAM on consumer hardware is a massive feat and a pretty cool as a statement for Strata. If your plan is to use it as a high-speed pair programmer where you'll do the code reviews, it might be fast and somewhat capable. But for unattended autonomous builds, sub-2-bit quantization loses the fine coherence needed for multi-turn state machines and you'd be better of with Qwen3.8-27B. Even UD-Q4_K_S did way better than Strata's Qwen3.8-Flash-Next Coder (IQ1_M - 125B MoE).

🎂Celebrating #XEN's 4th BIRTHDAY!🎂 On #Ethereum since Oct 8, 2022: 🔹 72.1M cRanks 🔹 33.2M active minters 🔹 18.9K active stakes 🔹 168.35T XEN staked 🔹 433.20T XEN supply 🔹 Currently at AMP 1,540, 4% APY & 491d mint term

#X1 Investors: Vesting claims open TODAY at 19:13 UTC at vesting.x1.xyz

$XEN on #X1 soon!

Testnet Airdrop day is HERE! 🔥 All validators that kept the #X1 testnet alive for months are getting rewarded today. Go ahead and check your wallets, and see your payout at testnet-rewards.x1.xyz

I just queried 111,218 #Ethereum gas records covering the full calendar months Aug–Sep 2026 and Ethereum activity is going up again. Spike episodes ≥3 gwei: August • Spike days: 3 • Average: 0.173 gwei • Peak: 10.087 gwei September • Spike days: 7 • Average: 0.284 gwei • Peak: 14.402 gwei September’s average gwei is 64% above August and spike episodes doubled, from 6 to 12.

As of today, the leading single-account #XEN minter claimed 1,787,751 cRanks and minted 11.06 trillion #XEN, spending 1,524.94 ETH (~$4M) in claim costs. The account uses its own contract to claim cRanks. TOP 1,000 accounts created 54.9M cRanks and so far minted 391 trillion XEN by paying almost 24k ETH (~$64M). That also means that 76% of all XEN created so far, was minted by the TOP 1K minting accounts. The leading single accounts across other minting methods: • #XENFT: 198,839 cRanks created • @cointool: 890,663 cRanks • #MCT: 287,480 mints • #MXENFT: 749,990 mints For the latest numbers, the XEN.pub dashboard refreshes every 12 hours: xen.pub/xen-top_cranks_eth.p…

🚨 XEN jumped +50.44% over the last 2 days (current price: $0.000000006371).

#X1 Testnet Validators: 10 days to claim your neptune-privacy:native rewards!

Good to see people minting $XEN being in profit again. Both max-term and short-term minters are turning a profit using XENFTs, MXENFTs, Cointool, MCT and other $XEN minters. Not everyone, of course, mostly those who were smart or lucky to mint at low gwei and time their maturity dates well. Total supply of all mints claimed already and future (as of now) is around 875 Trillion XEN.

I'm building a solution that uses @typesafeai's Jev as its judge and I call #Jev through @OpenRouter. This week I tested whether @jaredpalmer's open Kev-9B, running locally on a single RTX 4090, could stand in for it and these are results from my first run,... so take them as early notes, not the final word... ✅ Flags: they agree. Across 30 test cases, #Kev made the same "flag this" call as Jev in 43 of 48 checks. ❌ Ranking: they don't. When asked which output is better, the two rankings barely correlate (0.23). Kev favors the safe, literal option and marks down the more creative one. From my own review, Jev gets this right and Kev doesn't. 🐢 Speed: 3,970 calls took about 80 min on the 4090, at $0. Jev on OpenRouter would do the same in about 12 min for roughly $0.30. 💡 Prompt shape matters more than the GPU. If you plan to use Kev, put the content in state and keep the questions short. In my tests, the same content went from 241s to 6.3s. My verdict for now: I've switched Kev off and run everything on Jev via OpenRouter again. I'll keep Kev in my back pocket as an offline option for yes/no checks, in case OpenRouter or the TypeSafe API goes down, or I run out of credits. Just my experience, in case it helps someone. #LocalAI #RTX4090 #OpenSource #LLM

Last 5 x @Muse invites left. Each unlocks 1 billion non-expiring free tokens from @Meta (roughly 750M words in tokens). You won't be running out anytime soon. Download Muse, open the app or web version, go to Settings, tap "Redeem a Code" and use code: O6LY7Q.

I run a few of my own AI agents... Hermes (@NousResearch) and @OpenClaw to handle server maintenance and also have a personal assistant that keeps track of my schedules, reminds me of BDs, etc. But that cost tokens and I also had to create a secure sandbox for each agent to isolate their work. Anyways, I just started testing @Muse which @Meta finally launched in Canada yesterday and so far I am impressed. Normally, aside from Llama and @WhatsApp I have literally zero interest in anything from @Meta, but the way they engineered Muse is real nice (other than a ridiculous yeti mascot). The agent (not LLM) lives inside air-gapped @Debian container (2 vCPUs, 8GB RAM, 8GB storage) with Sentinel FW & zero-trust Authd credential engine) and long-term memory is inside @PostgreSQL db (my fav). So basically it runs on your own dedicated, end-to-end encrypted VM. I would NOT recommend allowing it to connect to your servers or store anything sensitive there, until someone figures out how to BYOK, but this is great for remembering events, kids trainings, sending reminders, monitoring website uptime, and similar everyday tasks. It also feels significantly faster than Hermes and OpenClaw and I can keep tokens to more important things. If you want to try it out, here is my 48-hour link that gives you 1 billion free tokens, no expiry: Code: O6LY7Q muse.ai/join Btw. That's equivalent of 750 million words and going to last you for years for regular tasks on top of their credits.