I Paid for a Top-Tier Model. It Got Dumber and Marked Me
Codex GPT-5.5 clusters reasoning at 516; Claude Code secretly marks Chinese users. When cloud AI silently degrades, open-weight self-hosting is the hedge.
14 min read
Blog
All Posts →
Intelligence Is Starting to Be About Wealth
Fable 5 ran a 15-hour task for $420. After June 22 it leaves Coding Plan for API rates, a 10x cost jump: access to top intelligence is a wealth problem.
3 min read
Claude 4.8 Let Me Down, But It’s Not Just Claude’s Problem
Claude Opus 4.8 kept failing my research and engineering tasks, so I moved them to GPT 5.5. Top models rotate—NVIDIA, AMD, and Zhipu prove survival wins.
8 min read
Projects
Personal projects and works →
C++HIPROCmAMD Ryzen AIWindowsLLM InferenceOpenAI API
AIMA AMD395 Qwen3.6 35B Windows Engine — Native Windows Inference
Native Qwen3.6-35B-A3B BF16 inference on AMD Ryzen AI Max+ 395: 8K prefill 2,126 token/s, decode 30.55 token/s, 131K prefix-reuse scenario TTFT 8.59s.
C++OpenVINOIntel ArcLevel ZeroU4LLM InferenceOpenAI API
AIMA Intel PTL Qwen3.6 35B U4 Engine: Specialized Intel GPU Inference
Intel PTL/Arc B390 OpenVINO U4 engine for Qwen3.6-35B-A3B: prefill ≥1.48×, decode ≥1.59× vs stock GPU; 10,752 greedy tokens bit-for-bit identical.
C++HIPROCmAMD Ryzen AILLM InferenceOpenAI API
AIMA AMD395 Qwen3.6 35B Linux Engine — Native Inference Engine
Native BF16 engine for Qwen3.6-35B-A3B on AMD Ryzen AI Max+ 395. No Python/PyTorch/vLLM/system ROCm. 262K ctx, OpenAI API, SSE streaming, function calling.