NVDA 125.53 ▼14.01%GOOGL 265.29 ▼2.70%MSFT 484.52 ▼2.77%AMD 424.39 ▼13.25%INTC 51.28 ▼7.97%TSMC 318.71 ▼9.48%AMZN 246.82 ▼4.10%META 560.84 ▼5.93%AAPL 342.56 ▼1.90%PLTR 143.49 ▼6.63%
Markets at last close

Models

Kimi leads Chinese models as costs and open weights reshape choices

·1 min read

Moonshot AI’s Kimi K3.1 Pro leads Chinese models on the Respan Index as of November 19, 2026, scoring 71.5 and ranking ninth overall. It trails GPT-6.2 Astra by 3.8 points and is the only Chinese model in the overall top 10. DeepSeek V4.1 Flash 1017 ranks next among Chinese models at 59.3, followed by Xiaomi’s MiMo-V2.6-Pro at 56.0. US labs retain the highest overall positions, while most leading Chinese alternatives compete with less expensive US offerings rather than flagships.

Task-specific performance changes the comparison. Kimi’s coding score exceeds GPT-6.2 Astra’s, while DeepSeek’s agents score surpasses Claude Haiku 5.5’s. Kimi costs $7 per million output tokens against $40 for Astra; DeepSeek costs $0.37 against Haiku’s $0.50. These prices reflect the cheapest provider routes on Respan’s gateway after discounts, not necessarily direct API rates. Only half of the leading Chinese models cost less than their matched US alternatives, and provider routes can differ in context length, tool support and rate limits.

Downloadable weights provide another distinction: 18 of the 20 ranked Chinese models publish weights or run on a published checkpoint. That enables self-hosting or independent hosting, although custom licenses can impose commercial conditions. Data handling depends on the serving provider, and some customer contracts restrict Chinese-origin models regardless of hosting location. The index combines public and lab-reported benchmarks, with some category scores estimated rather than measured.

Originally reported by respan.aiRead the source →
Related coverage