China's AI Model Rush: Kimi K3 Leads the Pack

Within a month, Chinese AI labs released several large models including Kimi K3, Qwen3.8, DeepSeek V4, and GLM-5.3, approaching frontier performance. Researcher Yuchen Jiang notes coding models are converging, with 95% of tasks indistinguishable. Kimi K3 is the highlight, and DeepSeek V4 includes an official inference framework.

Confirmed

According to @chibop1 relaying a Reddit list, released models include Kimi K3-2.8T, Qwen3.8-2.4T, DeepSeek-V4-Pro-0813-1.6T, and GLM-5.3-743B, with massive parameter counts.
@YuchenjUW (University of Washington researcher Yuchen Jiang) notes that AI coding models are converging: for 95% of tasks, users cannot distinguish between GPT-5.6 Sol, Fable 5, Kimi K3, GLM-5.2, or Qwen 3.8 Max. His post also mentions GLM-5.3 released that day (08-15).
@haider1 states Kimi K3 trails only Fable 5 and 5.6 Sol; DeepSeek V4 Flash/Pro approaches frontier in agentic and terminal coding.
@casperhansen considers Kimi K3 the highlight, Qwen3.8 Max stable, and DeepSeek V4 fully released with an official inference framework; he emphasizes that a good inference framework is now a key part of model releases.
@demianai notes open-source models are getting bigger, faster, cheaper, with improved agentic abilities; he also shares the real-time model comparison tool WhatLLM.

Unconfirmed

When @demianai posted (08-13), Kimi K3 and DeepSeek V4 were predictions; subsequent posts confirm actual releases, so predictions largely materialized.
The claim that Chinese labs may surpass OpenAI/Anthropic is an opinion relayed by @haider1, depending on whether the latter slows down; not yet settled.

Why it matters

Four labs releasing large models within a month shows accelerated iteration in Chinese AI.
According to the researcher, mainstream coding models are hard to distinguish on most tasks, so selection may depend more on cost, ecosystem, and inference frameworks.
DeepSeek V4's official inference framework indicates that inference frameworks are becoming a new competitive dimension.

2026-08-13 ~ 2026-08-15 · 6 related posts

Primary sources

In Less Than a Month: Kimi K3, Qwen3.8, DeepSeek-V4-Pro, GLM-5.3 Released — chibop1 · 2026-08-14 22:57 (UTC+08:00)
AI coding models converge; smaller open models like GLM-5.3 accelerate shift — Yuchenj_UW · 2026-08-15 01:01 (UTC+08:00)
Chinese models surge: Kimi K3, DeepSeek V4, Qwen 3.8, GLM 5.3 rival frontier — haider1 · 2026-08-14 17:00 (UTC+08:00)

Open Models Boom: Anticipating Kimi K3, DeepSeek V4 and More — demian_ai · 2026-08-13
Recent Model Releases Highlight Kimi K3 and DeepSeek V4 with Its Own Harness — casper_hansen_ · 2026-08-14
[source] Chinese models surge: Kimi K3, DeepSeek V4, Qwen 3.8, GLM 5.3 rival frontier — haider1 · 2026-08-14
Chinese AI Labs Release New Models in a Month, Closing Gap on Frontier — haider1 · 2026-08-14
[source] In Less Than a Month: Kimi K3, Qwen3.8, DeepSeek-V4-Pro, GLM-5.3 Released — chibop1 · 2026-08-14
[source] AI coding models converge; smaller open models like GLM-5.3 accelerate shift — Yuchenj_UW · 2026-08-15