Nanbeige4.3-33B benchmarked: 59 tok/s single stream, ~1,060 tok/s across 64 concurrent runs
MaziyarPanahi · x · 2026-10-26
Developer MaziyarPanahi shared real-world throughput numbers for the open-weight Nanbeige4.3-33B model, with weights available on Hugging Face.
- About 59 tok/s on a single stream
- 55 tok/s per stream with 10 concurrent runs, as shown in his demo video
- Roughly 1,060 tok/s total throughput at 64 concurrent streams
- The 10 demo prompts were picked by how long each model thinks, so streams finish around the same time — not selected by correctness
More from Models
- Fintech insiders' AI consensus: ChatGPT, Claude and Grokbot top the pack — LexSokolin · 2026-10-26
- DeepSeek-adjacent researcher hints next model will keep the V3.2 DSA approach — YouJiacheng · 2026-10-26
- Anthropic S-1: 1,300% YoY revenue growth, 85% gross margin — danshipper · 2026-10-26
- EMNLP paper: tweaking multiple-choice scoring schemes reshuffles LLM leaderboards — boydgraber · 2026-10-26
- AI answers can cite real, working sources and still be flat-out wrong — alifcoder · 2026-10-26
- Kimi K3 needs ~1 TFLOPs per decoded token while V4.1F needs only 50 GFLOPs — YouJiacheng · 2026-10-26