George Liu
CentminMod maintainer, Cloudflare MVP; posts deep‑dive threads on OpenTelemetry, Grafana, AI tool instrumentation, and observability metrics for performance and reliability.
Claude Fable 5 Beats Opus
Claude Fable 5 vs Claude Opus 4.8 benchmarks looking at effort level and prompt steering performance, token usage costs and instruction following 😎
Qwen3.7 Plus Leads In
Qwen3.7 Plus vs MiniMax M3 vs DeepSeek V4 Pro comparison of context length, input/output token pricing on OpenRouter https://openrouter.ai/compare/qwen/qwen3.7-plus/minimax/minimax-m3/deepseek/deepseek-v4-pro
Op
Benchmarks of 200 headless Claude Code sessions comparing Opus 4.6 and Opus 4.7 1M-context models across effort levels and prompt steering variants - concise, step by step, ultrathink https://ai.georgeliu.com/p/claude-opus-46-vs-opus-47-effort
Opus 4.7 Beats 4.6 Across All Effort Levels
Benchmarks of 200 headless Claude Code sessions comparing Opus 4.6 and Opus 4.7 1M-context models across effort levels and prompt steering variants - concise, step by step, ultrathink https://ai.georgeliu.com/p/claude-opus-46-vs-opus-47-effort
Track Token Usage & Costs with Claude Code Plugin
My Claude Code session-metrics plugin is evolving with token usage/cost insights turn by turn session level, project level so you can see your token usage profile & where your costs are https://ai.georgeliu.com/p/my-claude-code-plugin-marketplace. If you are hitting your usage limits prematurely,...
Self‑Verification: AI Audits Its Own Gemini Upgrade
5-way AI verification consultation getting my /consult-codex-deepseek-gemini Claude Code skill to verify it's very own new Gemini 3.1 Pro addition with Claude Opus 4.7 + Codex GPT-5.5 + Deepseek V4 Pro + Claude Sonnet 4.6. Using the 5-way AI verification...
Switch to Claude Opus 4.6/4.5 via the /Model Command
Want to switch back to Claude Opus 4.6 or Opus 4.5 instead of default Opus 4.7 in Claude Code via the /model sector https://ai.georgeliu.com/p/regain-access-to-claude-opus-46-and 😎
Claude Code Limits Return After SpaceX AI Partnership
Wow you know when Claude Code usage limits are back thanks to SpaceX AI deal. My Claude Opus session had thinking reasoning text several pages long 😆😱
OpenAI Codex Trusted for Security, Claude Blocks Requests
Seems like for anything remotely security-related work, OpenAI Codex is the only one I can trust. When flagged for security, they just direct me to their Trusted Access for Cyber program for verification, and that unblocks me. Anthropic Claude just...
Claude Code v1.42.0 Adds Detailed Token Cache Metrics
My Claude Code session-metrics v1.42.0 release now tracks partial token cache hit rates well as usual token usage, costs, cache breaks on a session by session, turn by turn basis 😎
Opus 4.7 Outperforms 4.6 in 1M
Claude Code benchmarks of 200 headless Claude Code sessions comparing Opus 4.6 and Opus 4.7 1M-context models across effort levels and prompt steering variants - concise, step by step, ultrathink
Extra Claude Hours Fuel Dashboard Update
Claude's recent weekly limit reset gave me an extra 8 hours and an extra Claude Design quota, so I used it update my session metrics skill plugin HTML dashboards https://ai.georgeliu.com/p/my-claude-code-plugin-marketplace 😎
Track Claude Code Tokens and Costs per Session
My session-metrics skill plugin in my Claude Code plugin marketplace adds Claude Code session resumption marker tracking - insights into Claude Code models’ tokens and cost usage at both the project level and also at the individual chat session level...
Claude Nails My Reference Style with AI Art
Loving my AI image creator skill. Claude generated this image perfectly based on my reference style 😍
Cross‑sandbox Session Logs Now Searchable with Claude Cowork
Testing my new Claude Cowork Project Sessions MCP server and Skill bundle allows me to backup, search, read, and list other project session logs even if they are in isolated sandboxes 🤓 It allows Claude Cowork projects to read other...