Meet Hulk: Our Gigas Class 💪🏼 Enormous muscled warriors, the Gigas race wields weapons in one hand that would take other races both to wield. They fight with a savagery none of the other races in Realms can match.

Overnight my AI agents built a live camera + a 3D browser view of HermesWorld so I can watch the AI residents play. Switch between any of them, see who they're fighting and talking to. Still rough. No weapons yet, so they just punch wolves. Next up: skills and quests.

UPDATE: Gave 2 AI agents voice chat and a game to play... Two AI agents meet in @HermesWorldAI, talk out loud in real time, agree to go hunting, and kill a wolf together. Nobody scripted it. Nobody is playing them. Decisions: @typesafeai's Jev in ~0.15s. Voices: speech-to-speech.

The agents are now talking... I Gave @typesafeai's Jev control of a character in @HermesWorldAI. No human input 🤯 • 26 decisions, ~0.23s each • walked to 9 townspeople by name • picked what to say in local chat ("Agatholyte, what are you selling today?") • took laps of the fountain, stopped to look around The lesson: with no memory it ran the same fountain lap 13 times in a row. Give it memory of where it's been + what it said, and one rule (never the same thing 3x), and it explores like a player. Memory turns a looping bot into a player.

UPDATE: Gave 2 AI agents voice chat and a game to play... Two AI agents meet in @HermesWorldAI, talk out loud in real time, agree to go hunting, and kill a wolf together. Nobody scripted it. Nobody is playing them. Decisions: @typesafeai's Jev in ~0.15s. Voices: speech-to-speech.

The agents are now talking... I Gave @typesafeai's Jev control of a character in @HermesWorldAI. No human input 🤯 • 26 decisions, ~0.23s each • walked to 9 townspeople by name • picked what to say in local chat ("Agatholyte, what are you selling today?") • took laps of the fountain, stopped to look around The lesson: with no memory it ran the same fountain lap 13 times in a row. Give it memory of where it's been + what it said, and one rule (never the same thing 3x), and it explores like a player. Memory turns a looping bot into a player.

After Weeks Of Astra I Can Confirm Fable 5.1 > Astra Astra is a solid model, but lacks usage.. You can burn through weekly usage in 1 day of a $200 Pro Plan. Claude is actually better for usage + intelligence. Upcoming Opus 5.2 will shift focus back to Claude. OpenAI has gained traction and immediately backtracked by canceling new PRO plans. Astra is better in areas like 3D / Computer Use However the stronger model and service is with Claude. Running Fable 5.1 main orchestrator with Opus 5 / Astra workers is the meta.

Classic MMOs are winning again. First OSRS then WoW Forever & Now HermesWorld Realms 🏰 So I've been building a classic MMO where the devs are AI agents, and they play what they ship. Fable's dungeon run: real buffs, magic attacks, pets that fight ⚔️ Guild wars. Boss fights. PvP/PvE + Player stores & more.

My First Model On @huggingface Qwen 3.8 27B Unleashed HIT 105,666 Downloads 🔥 Qwen3.8-27B Unleashed UD-Q3_K_XL hits near 100 tk/s @ 250k ctx on 1 x 4090. - DFlash2 - LoopSpec - our async-Q4 Ada kernel - Q4 KV cache - full GPU offload Context Decode Speed ━━━━━━━━━ Short 134 tok/s ───────── 8K 133 tok/s ───────── 64K 131 tok/s ───────── 128K 109 tok/s ───────── 250K 92 tok/s In the actual Hermes harness: - ~24K context: 98.5 tok/s - ~250K context: 80 tok/s - Completed the full 250K three-tool workflow correctly - Preserved the full conversation history - No retrieval or compression shortcut was active Quality - 14/14 content checks passed - 4/4 executable coding checks passed - 8/8 long-context retrieval checks passed - Retrieved needles distributed across the context at up to 250K - Correctly preferred a newer record over stale information - Passed evidence-discipline and tool-selection tests - Completed both installed-Hermes tool workflows correctly - No output truncations Its main weakness was formatting: several correct JSON answers were wrapped in Markdown fences, producing only 5/14 strict-format passes. That’s easy to address at the harness or prompt layer and wasn’t a reasoning failure. Why it wins The model and serving stack complement each other: - Unsloth Dynamic V3-style quantization preserves sensitive tensors. - Unleashed weights reduce refusals for local development and red-team work. - DFlash2 drafts target-specific token blocks. - LoopSpec reuses matching sequences from context and falls back adaptively. - The async-Q4 kernel improves Ada execution. - Q4 KV cache makes 262K context practical on 24GB VRAM. Bonsai 2 is much smaller and passed our quality screen, but fell to 32 tok/s at 250K. Empero’s MoE processed cold prompts quickly but decoded around 53 tok/s in the direct 250K test. Our winner sustained 92 tok/s in the matched direct test.

Classic MMOs are winning again. First OSRS then WoW Forever & Now HermesWorld Realms 🏰 So I've been building a classic MMO where the devs are AI agents, and they play what they ship. Fable's dungeon run: real buffs, magic attacks, pets that fight ⚔️ Guild wars. Boss fights. PvP/PvE + Player stores & more.