The complete archive
7 published articles, newest first.
3 PDF Parsers Tested: Docling vs Marker vs PaddleOCR
Docling vs Marker vs PaddleOCR on a 24GB unified memory machine. Marker is 30x faster than PaddleOCR. Real numbers from 12 runs across 4 PD…
Q4 vs MXFP4: Which Quant Is Faster?
Q4_K_M is 1.8x faster than MXFP4. Same code prompt, 3 trials each, 6 total runs. gpt-oss-20B vs Qwen3-14B. Hard numbers, no vibes.
4 Free LLM APIs vs Claude: 5 Coding Tasks, Real Data
3 free LLM APIs (NVIDIA NIM, Groq, OpenRouter) vs paid Claude on 5 real coding tasks. All 20 runs passed. Speed varied 10x. Hard numbers, n…
gpt-oss-20B vs Qwen3-14B vs Mistral-Small-24B: A Real Benchmark
Three Apache 2.0 open-weight 14-24B LLMs head-to-head: gpt-oss-20B, Qwen3-14B, Mistral-Small-24B. MMLU, GSM8K, HumanEval, IFEval benchmarks…
MLX vs llama.cpp on Apple Silicon: Same Model, Two Engines
On Apple M2 with 24 GB, MLX is 1.42x faster than Ollama on chat, but Ollama is 1.7x faster on RAG. Real wall-time numbers, same Qwen3 14B 4…
Local LLM Tool Calling for AI Agents: Qwen3 14B vs Llama 3.2 3B on Apple Silicon
Qwen3 14B picks the right tool 9/10; Llama 3.2 3B is 25x faster but only 5/10. Real M2 agent benchmark, what to ship for tool calling on a…
Qwen3 14B vs Llama 3.2 3B on Apple Silicon M2: An Honest Benchmark
Qwen3 14B vs Llama 3.2 3B on Apple M2 24GB: 3.5x faster throughput, 4x less memory — but the 14B hits a token-budget wall. Real numbers and…