Kimi K3.1 Pro hits 0.78x human compute multiplier, lagging frontier by ~4.6 months
ChowdhuryNeil · x · 2026-11-05
pzeroresearch evaluated Kimi K3.1 Pro on TasteVal, finding an experiment compute multiplier of 0.78x relative to their human experts — roughly 4.6 months behind the frontier and on par with Opus 4.8.
They also fit a separate trend for leading open-weight models: on average these lag the frontier by about 4.2 months, with doubling times similar to frontier models. The chart shows frontier and open models (Kimi K-series, GLM 5.3) converging toward human expert level along the same exponential curve from GPT-4 to Opus 5.5.
Related event: Kimi K3.1 Pro Scores 0.78x Compute Multiplier in TasteVal Benchmark(2 posts)→
More from Models
- Report: 80-90% of OpenAI research is focused on GPT-7 and beyond — VraserX · 2026-11-05
- Early demos show Kimi K3.1 Pro generating 3D worlds directly — _akhaliq · 2026-11-05
- Ripple-65B team adds hardware verification lead Caroline Trippel amid launch — annadgoldie · 2026-11-05
- Adtech veteran outlines the personalization blueprint: segments, features, real-time routing — jonathanmendez · 2026-11-05
- Claude trains a model instead of writing code: 1257-param pixel-art generator in 10 minutes — victormustar · 2026-11-05
- MoE beats dense in matched OOD finetune: DeepSeek V3.1 hits 0.804 CE vs Llama 405B — kalomaze · 2026-11-05