Best sub-40B model on Terminal-Bench 4.0: Nanbeige4.3-33B at 23.8% accuracy
NielsRogge · x · 2026-10-26
A lookup of Terminal-Bench 4.0 shows the only listed model under 40B parameters is Nanbeige4.3-33B (33.1B), scoring 23.8% accuracy with MiniSWE-Agent — the sole qualifying entry on the leaderboard.
More from Models
- LlamaIndex releases OCRia-1, topping ParseBench at 87.4 with 2-4x faster decoding — llama_index · 2026-10-26
- "90% of tokens local, 10% frontier": HF engineer's model allocation take — TheZachMueller · 2026-10-26
- Iranian users report Qwen chat access blocked; blogger says sanctions aren't Alibaba's call — dotey · 2026-10-26
- Hugging Face releases Carbon-X, an open 14.8B DNA language model that reads genomes in chat — _lewtun · 2026-10-26
- Users annoyed as ChatGPT keeps lecturing them with "Kevin, ..." openers — kevinkern · 2026-10-26
- Frontier models are better judged by five minutes of weird jokes than 40 benchmark numbers — korymath · 2026-10-26