vLLM Highlights TPU Work Powering AgentX Results Across Three Models

Who: vLLM
What: highlighted the vLLM ecosystem’s TPU work, including the Kimi K3 megakernel and vLLM TorchTPU powering AgentX results across GLM-5.2, Qwen3.5 397B, and Kimi K3
Quote: Thanks to @SemiAnalysis_ for highlighting the vLLM ecosystem’s work on TPU, in particular the Kimi K3 megakernel and vLLM TorchTPU powering AgentX results across GLM-5.2, Qwen3.5 397B, and Kimi K3.

What happened

Thanks to @SemiAnalysis_ for highlighting the vLLM ecosystem’s work on TPU, in particular the Kimi K3 megakernel and vLLM TorchTPU powering AgentX results across GLM-5.2, Qwen3.5 397B, and Kimi K3. Shoutout to @inferact and @googlecloud for pushing vLLM performance on TPU. The

Links in the post: 1

Article

Thanks to @SemiAnalysis_ for highlighting the vLLM ecosystem’s work on TPU, in particular the Kimi K3 megakernel and vLLM TorchTPU powering AgentX results across GLM-5.2, Qwen3.5 397B, and Kimi K3.

Shoutout to @inferact and @googlecloud for pushing vLLM performance on TPU. The https://t.co/Iut0KHJHKD

Quoted postSemiAnalysis@SemiAnalysis_

The Iron(wood) Throne: Google's TPU Stakes Its Claim on Inference Viable Alternative of CUDA Moat, NVIDIA Working Nights & Weekends, Jensen’s Counter Submission, Kimi K3, Meta Muse Spark Open, GLM5.3, TorchTPU, SGLang, vLLM, llm-d, Ray https://t.co/mh16HIWmpJ

View the quoted post on X

The story so far

vLLM highlighted TPU work from its ecosystem behind AgentX results for GLM-5.2, Qwen3.5 397B, and Kimi K3. It pointed to the Kimi K3 megakernel and vLLM TorchTPU, and credited Inferact and Google Cloud with pushing vLLM performance on TPU. vLLM said SemiAnalysis had highlighted this work; the post does not specify the results or performance figures.

Model releases, price changes and compute developments, with original sources. Free, confirmed by email, and easy to unsubscribe. Read a past issue · 2026-11-05

vLLM Highlights TPU Work Powering AgentX Results Across Three Models

What happened

Read next

The story so far

Keep up with AI, one email a day