Tencent Hy4 is open-weight. Its 49B-active label still needs 214 GiB
Tencent Hy4 activates 49B parameters, but its smallest official low-bit build is 213.66 GiB and needs patched inference software.
DeepSeek, Kimi, Qwen, and Chinese models reshaping the race.
Tencent Hy4 activates 49B parameters, but its smallest official low-bit build is 213.66 GiB and needs patched inference software.
Qwen3.8-Flash-Next activates about 6B parameters per token but carries far more weights. Here is how to test memory, context, speed, cost and licensing.
DeepSeek V4 Pro 0813 is live across app, web, and API. New peak and off-peak rates begin August 16 and change the routing math.
DeepSeek Flash 1017 is now on Workers AI. Compare provider prices, access requirements and rate limits separately from native API pricing and Fireworks training.
Chinese open models now hold most of the top open-weight slots and match the closed frontier on some coding tests. Here's how to evaluate…
Kimi K3 is now available on Amazon Bedrock with vision, a one-million-token context window, prompt caching and cross-region inference profiles.
A plain-English buyer's map of China's big four open models - Kimi, DeepSeek, Qwen, GLM - what each is best at in 2026, and…
US companies are quietly running Chinese open models - because they're capable, cheap, and self-hostable. Here's why it's happening and what it means for…
From DeepSeek's shock breakout to Kimi's frontier push, China's open-model wave stopped being a footnote. Here's the story arc - and why the competition…
Alibaba put Qwen AI inside Taobao's whole catalog - agentic shopping, virtual try-on, hundreds of millions of users. This is what it looks like…