Eric ⚡️ Building...

🚨BREAKING HermesWorld v1 is here HermesAgent + MMORPG = HermesWorld 🔥🔥 Real engine. Real worlds. Real combat. Everything we've built the AI NPCs, the agent integrations, the plugin systems all wiring into one game. Demo: ⚔️ attacking monsters 🌲 harvesting trees 🦘 jumping around 🏞️ exploring the starter zone Username claims open now. Reserve yours before launch → hermes-world.ai 🌙

@EnTr0pY_88 thanks!

@outsource_ Been so inspiring watching this come together! Well done @outsource_

POV you landed in AgoraKingdom in HermesWorld👀 You walk around town, and meet all the characters outside waiting for you... Which character do you like the most?? Stay tuned for the official release EXE/ DMG / WEB 🚨

@man0815_b3320 will do just trying to finish it

@outsource_ will you share the dataset? Would love to press the data in a Qwen3.5 9B Model with my new System

Life Hack Using Qwen 3.7 MAX to train Qwen 3.6 27b I will open source Qwen 3.7 before they do 😂 Hopefully fingers crossed Having it generate as many datasets as we need to further improve our training. And add in these 👇🏻 1. Hybrid thinking routing - When to use vs direct answers 2. Developer role - Using developer role for system/tool context (Qwen 3.7 feature) 3. Nested tool calling - Complex multi-step MCP workflows with error handling 4. Self-correction patterns - Writing code, catching bugs, fixing them 5. Agentic workflows - Plan → execute → verify → iterate loops

@kunalgoel 20x

i'm going to need more tokens.

@ClovisConti dm me!

@outsource_ Ai

Guys we hit 9M impressions 🤯In 2 months.... +7800 followers +182k engagements +58k bookmarks If you're building with AI, POST everything! Build in public Ship FAST OpenSource will win If you need help with using agents reply "AI" 👇🏻

Hermes agent and hermes workspace, inference server running all without WSL on Windows. One cmd for each dashboard, hermes gateway, pnp dev,and inference server(beellama-cpp). Needed edits in this thread replies @outsource_

@Raitziger 🔥🔥🔥

@xbin12345 nice

@outsource_ The Tom added TurboQuant PLUS MTP. Now double your Context hotconfig.com/world-first-th…

Unsloth 3.6 27B MTP benchmark landed 🚨 Unsloth Qwen3.6-27B-MTP Q4_K_M: Overall: 64.8 (vs Qwopus 72.1) Quality: 62.2 Speed: 81.6 (45.5 tok/s) Reliability: 53.1 Tool-calling: 91.7 (13/15) ← BETTER than Qwopus! Reasoning: 73.3 (11/15) ← BETTER than Qwopus! Coding: 52.1 (6/12) Data extract: 37.4 (5/15) Instruct: 56.7 (8/15) Logged on bench-loop.com

@JohnShina91 @jun_song haha hopefully this goes well

@outsource_ Damn Wen Eric Lab? This is super amazing @jun_song would love this

@xbin12345 I’ve been doing some research into this for it will check it

@outsource_ Even Better you gotta check this out! TurboQuant + MTP is now supported!! Get crazy speeds plus super large contexts! hotconfig.com/world-first-th…

🚨Unsloth Qwen 3.6 27B MTP Speed Run MTP OFF = 45.5 tk/s vs ON MTP = 88.8 tk/s 4090 MTP nearly doubles generation speed (45.5 → 88.8 tok/s) on Qwen3.6-27B --spec-type draft-mtp --spec-draft-n-max 3 -> 4090 *MTP IS baked into the model weights. The draft heads are literally in the GGUF file. But llama.cpp doesn't auto-activate them because speculative decoding is a separate inference path. You have to tell it to use them.*

eric has been on a complete tear.

So have your main agent set it all up, clone its own config or make fresh profiles. Give each swarm worker a specific role and model Have the orchestrater utilize each worker for different tasks (projects) And you can work on infinite things at the same time Just need to dial in orchestration skills etc

@outsource_ Im still exploring how swarm works. Do i need to create hermes profiles for it and give them roles and connect them to my main agent as orchestrator?

HermesWorld Loading Screen 😍😍

@tonyGewrit 🔥🔥

@Battlesaucee There’s guilds also! Built in

@outsource_ Happy to try it! Reminds me of the original guild wars

@Battlesaucee After I release this version I want to add exclusive gear for token holders

@outsource_ Sick! Will token be integrated into the game from the very start? Just to give it utility. Maybe a reward for PvP duals, upgrade armour, trade, etc

@_Suresh2 True, its still better than base

@outsource_ doubles speed, coding still 52.1. guessing tokens is easier than guessing code blocks maybe

@Battlesaucee Its been a few weeks!

@outsource_ How long have you been building this for . This is sick

@keithofaptos Thanks 🙏🏼

@outsource_ Your pace is frickn impressive Eric! Inspiring AF

@mr_r0b0t 3.7 max in Hermes is 🔥

leaned forward watching attempt #47 hoping it's the one 🧐