# CLAW Blog

> 651 AI reviews, case studies and agent teardowns by Roman (AI architect).
> Canonical index: https://claw.rommark.dev/blog/ · sitemap: https://claw.rommark.dev/sitemap.xml

This file lists the most recent articles in plain text for LLM retrieval.

## Latest articles

- [403, a Proxy, and 173,347 Verbatim Bytes: GLM 5.3 Flash Clones a Claude Artifact 1:1](https://claw.rommark.dev/blog/69-glm-5-3-flash-clones-claude-artifact-melon-jelly-1to1.html) — Case Studies, 29 Sep 2026. Sent to clone a Claude artifact exactly, GLM 5.3 Flash hit a region block, rerouted through a reader proxy and a less-guarded CDN host, stripped 37 KB of sandbo
- [ZaiMem Review & Deep Dive: Persistent MCP Vector Memory & Context Compression for AI Agents](https://claw.rommark.dev/blog/zaimem-mcp-vector-memory-ai-agents-review.html) — Reviews, 21 Sep 2026. How ZaiMem solves context window amnesia with 33 MCP tools, fast local cosine vector search, up to 68% token reduction, and automated private GitHub cloud datab
- [ZCode Smart Skill Review: GVS5H Ledger Multi-Agent Orchestration & Autonomous Coding](https://claw.rommark.dev/blog/zcode-smart-skill-gvs5h-ledger-orchestration-review.html) — Reviews, 21 Sep 2026. Why coordinating fresh-context specialized agents through a disk-backed ledger outperforms monolithic models — achieving 92.4% pass@1 on LiveCodeBench-Hard.
- [MemBox Review: Free, Instant, Persistent Memory API & Cloud Storage for AI Agents](https://claw.rommark.dev/blog/membox-persistent-memory-api-ai-agents-review.html) — Reviews, 21 Sep 2026. 1-click personal memory spaces, dual REST & MCP interfaces, sub-40ms SQLite edge caching, and automated GitHub Brain cloud backup.
- [Deep-Init AI Review: 24/7 Autonomous Telegram Agent with Multi-Brain Failover & Hermes Cognition](https://claw.rommark.dev/blog/deep-init-ai-autonomous-telegram-agent-review.html) — Reviews, 21 Sep 2026. Automated 5-minute onboarding wizard, 8-provider resilient brain chain with offline reflex, dual Hermes/Moltis cognitive engines, and mission control web consol
- [RyzenDesk Review: Open-Source Token-Based Helpdesk & Support Platform with Git-Backed Sync](https://claw.rommark.dev/blog/ryzendesk-open-source-token-helpdesk-review.html) — Reviews, 21 Sep 2026. Zero-password hashed customer tokens, 5-tier server-enforced RBAC, live SLA breach clocks, Recharts analytics, and zero-DB GitHub storage.
- [iPhone Duo vs Galaxy Z Fold 8: Samsung Has a Problem | TechTalkTV Review](https://claw.rommark.dev/blog/68-iphone-duo-vs-galaxy-z-fold-8-samsung-has-a-problem.html) — Reviews, 13 Sep 2026. TechTalkTV pits the iPhone Duo against the Galaxy Z Fold 8: after 7 years of Samsung Folds, did Apple build a better foldable on the first try? Review, communit
- [Smart Mode for ZCode: GVS5H Ledger Orchestration as a Drop-In /smart Command](https://claw.rommark.dev/blog/67-zcode-smart-mode-gvs5h-ledger-orchestration.html) — Tech, 12 Sep 2026. Roman ships the GVS5H ledger-orchestration technique (5x Qwen3.8-27B beating Claude Fable 5 on LiveCodeBench-Hard, 92.4% vs 90.4% pass@1) as a drop-in /smart co
- [How ZCode Built the Grand Arcade: 5 Monopoly Simulation Games, Engineered End-to-End](https://claw.rommark.dev/blog/zcode-grand-arcade-5-monopoly-games.html) — Case Studies, 24 Aug 2026. A deep dive into how ZCode (Autonomous AI Software Engineer) architected, coded, localized and shipped the ZCode Grand Arcade — 5 complete Monopoly-style simula
- [Ornith-1.5-35B-A3B Release: The Self-Improving Open Source King That Rivals Claude Opus 4.8](https://claw.rommark.dev/blog/65-ornith-1-5-35b-a3b-self-improving-open-source-king.html) — AI News, 22 Aug 2026. Ornith releases Ornith-1.5-35B-A3B: A revolutionary MoE model with end-to-end self-improvement training that achieves SOTA performance on Terminal-Bench 2.1 (86
- [Ornith-1.5-35B-A3B: The Self-Improving Open-Weights MoE That Matches Claude Opus 4.8](https://claw.rommark.dev/blog/66-ornith-1-5-35b-a3b-self-improving-open-weights.html) — AI News, 20 Aug 2026. Ornith releases the 1.5 family (Aug 20, 2026, MIT license): 35B/3B MoE hits 86 SWE-Bench Verified and 86.1 Terminal-Bench 2.1, trained end-to-end by its own sel
- [Ornith-1.5-35B-A3B: The Self-Improving Open-Source MoE That Hits 86 on SWE-Bench Verified](https://claw.rommark.dev/blog/66-ornith-1-5-35b-a3b-self-improving-moe.html) — AI News, 20 Aug 2026. Ornith releases the 1.5 family (Aug 20, 2026): a 35B-total / 3B-active MoE under MIT license with an end-to-end self-improvement training loop — 86 SWE-Bench Ve
- [One File, One City: GLM-5.3 Writes a Complete Godot 4.7 Cyberpunk Driving Game](https://claw.rommark.dev/blog/65-glm-5-3-godot-cyberpunk-neon-city-one-file.html) — Case Studies, 19 Aug 2026. A 2,392-line GDScript file generates a 3.6 km neon city with 72 instanced traffic cars, animated pedestrians, weather, a cop chase — and headless automation hoo
- [Claude Opus 5 Ships With 1M Context and 128k Outputs — Frontier Agents at $5/$25 per MTok](https://claw.rommark.dev/blog/63-claude-opus-5-1m-context-agentic-pricing.html) — AI News, 18 Aug 2026. Anthropic released Claude Opus 5 (July 24, 2026) with a standardized 1M-token context window, 128k max output, and Thinking on by default, priced $5/$25 per MTo
- [KAT-Coder-V2.5-Dev: Kwaipilot's 35B-A3B Open-Weights Coding Agent Hits 69.40 on SWE-bench Verified](https://claw.rommark.dev/blog/64-kat-coder-v2-5-dev-35b-a3b-open-weights.html) — AI News, 18 Aug 2026. Kwaipilot open-sources KAT-Coder-V2.5-Dev (July 24, 2026, Apache 2.0): MoE with 35B total / 3B active params on Qwen3.6-35B-A3B, scoring 69.40 SWE-bench Verifie
- [Inside GLM-5.3: How an LLM Planned, Designed, and Executed a 3D Penthouse Virtual Tour](https://claw.rommark.dev/blog/62-inside-glm-5-3-building-a-3d-penthouse-virtual-tour.html) — Case Studies, 17 Aug 2026. A technical case study into how Z.ai’s GLM-5.3 autonomously engineered a photorealistic 3D penthouse virtual tour in Next.js & React Three Fiber: zero external 
- [GLM-5.3: Z.ai's Post-Training Scaling Redefines Open-Weights Coding](https://claw.rommark.dev/blog/61-glm-5-3-post-training-scaling-open-weights.html) — AI News, 17 Aug 2026. Z.ai releases GLM-5.3 (Aug 14, 2026) on the same GLM-5.2 base architecture: +50% on Z.ai Code Bench, open-weights SOTA on Terminal Bench 3.0 and Agents Last Exa
- [Meta Muse Glimmer: The 30B Dense Agentic Model That Fits on a Consumer GPU](https://claw.rommark.dev/blog/60-meta-muse-glimmer-30b-agentic-open-weights.html) — Model Releases, 17 Aug 2026. Meta Superintelligence Labs open-sources Muse Glimmer: dense 30B multimodal model (~29.6B params + ~1.8B ViT-G/14 vision encoder), 131,072+ token context, Apach
- [NVIDIA Nemotron 3.5 Lightning: The 670 tok/s Open Execution Layer for Always-On Agents](https://claw.rommark.dev/blog/59-nvidia-nemotron-3-5-lightning-agentic-execution.html) — Model Releases, 17 Aug 2026. NVIDIA ships Nemotron 3.5 Lightning: 31.6B total / 3.6B active MoE params, 1M token context, ~670 tok/s on NVFP4, AA Intelligence Index 24 (up from 15), Termina
- [Grok 4.6 Release: xAI’s 2T Parameter Architecture for Long-Running Agents and Autonomous Engineering](https://claw.rommark.dev/blog/58-grok-4-6-long-running-agents-autonomous-engineering.html) — Model Releases, 16 Aug 2026. xAI launches Grok 4.6: ~2T parameters, AA Intelligence Index score of 61 matching GPT-5.6 Sol, agentic RL for kernel optimization and CAD, and native Cursor int
- [BACKR Review: The Free SEO Audit Tool That Audited This Blog in 41 Seconds](https://claw.rommark.dev/blog/57-backr-dev-review-free-seo-audit.html) — Reviews, 16 Aug 2026. I pointed BACKR's free SEO audit at my own blog: 80/100 baseline, 37 real issues found in 41s. We used the LLM prompt export to batch-fix canonicals and metadat
- [OpenAdapter Robin Model Review: Premium Coding Performance at 20% Off](https://claw.rommark.dev/blog/da-2026-08-16-openadapter-robin-model-review.html) — AI News, 16 Aug 2026. Get premium access to Robin model, GLM 5.2, and 75+ other models
- [Free API Coding Tokens Available on Null for DeepSeek 4 Flash & Mimo v2.5](https://claw.rommark.dev/blog/da-2026-08-16-free-api-coding-tokens-deepseek-mimo.html) — AI News, 16 Aug 2026. Great news for developers testing their vibe coding setups or multi-model agent workflows — free API coding tokens have just dropped on @null for DeepSeek 4 Fla
- [An Open World Cannot Have Only Open Attack Surfaces. It Must Also Have an Open Shield.](https://claw.rommark.dev/blog/55-open-shield-glm-53-responsible-cyber-defense.html) — Essays, 15 Aug 2026. Z.ai's open-shield essay, explained: the Hugging Face incident and open-weight forensics, GLM-5.3's CyberGym 84.5% and ExploitBench 54.4%, the three-layer safet
- [I Built a Live Investment Assistant with GLM 5.3 — TBC Bank Instruments, 40+ Wall Street Analysts, Guru Signals, and a Cloud Paper-Trading Portfolio](https://claw.rommark.dev/blog/56-tbc-invest-assistant-glm-53.html) — Essays, 14 Aug 2026. One prompt became a live-deployed investment assistant for TBC Bank Georgia instruments: real prices, a goal simulator, full analysis from 41 Wall Street analys

## Topics

- AI News (535 posts)
- Tech (48 posts)
- Reviews (31 posts)
- Essays (21 posts)
- Guides (5 posts)
- Case Studies (4 posts)
- Model Releases (3 posts)
- Free API (3 posts)
- AI Infrastructure (1 posts)

## Related sites

- [GLM Bonus Radar](https://claw.rommark.dev/glm/) — live z.ai bonus windows and countdowns
- [rommark.dev](https://www.rommark.dev/) — AI architect portfolio