Nanbeige 4.3-33B flops real-world test: zero compilable output vs Qwen
julianharris · x · 2026-10-26
A fair head-to-head local-model test (identical harness, rebuild a real CLI tool, 502 hidden tests, no internet, 30 steps max) shows Qwen 3.8-27B on vLLM NVFP4 producing 251 off / 253 low / 308 medium tier results, while Nanbeige 4.3-33B on Q4KM llama.cpp generated literally nothing that would compile.
Julian Harris concludes Nanbeige's claimed benchmark gains are extreme benchmaxxing — trustmebro numbers — and has crossed it off his local model list.
More from Models
- Ruff creator Charlie Marsh: frontier models find infinite bugs in any software — charliermarsh · 2026-10-26
- Dev testing one skill across Opus 5.5 and Sol 6.1 finds instruction following 'all over the map' — lucasmeijer · 2026-10-26
- supermemory unveils learner-1, claiming continual learning for any model — julianweisser · 2026-10-26
- Anthropic says Opus 5.5 cuts cost per task by ~40% vs Opus 5 — taufiqintech · 2026-10-26
- AI search APIs ranked: top 12 within 7 points, yet prices differ nearly 7x — FinanceYF5 · 2026-10-26
- OpenAI Dots coming to EU/EEA and UK in a few weeks, slides confirm — testingcatalog · 2026-10-26