MiniMax M3.1 Flash follow-up: Baidu-affiliated chipmaker Kunlunxin says the model runs on its accelerators, as Chinese chips race for Day 0 support (report)

Verified (fact)

According to AI Times Korea (인공지능신문), Kunlunxin, which began as Baidu's AI chip unit, said on the 23rd (local time) that it had run MiniMax's new LLM 'M3.1 Flash' on its AI accelerator platform. The report presented this as an example of 'Day 0' compatibility, meaning support immediately after a model's release. The excerpt does not include performance figures such as inference speed or the chip generation used.

Needs fact-check (claim)

Kunlunxin reportedly says it aims to support new models as soon as they are released to broaden the use of Chinese-made AI chips.

Why it matters (interpretation)

As more Chinese open models run on Chinese chips right after release, the case grows that models can be deployed in China without NVIDIA GPUs. Without performance figures, however, it is still hard to judge whether this is good enough for production services.

Sources

• Press 중국 AI 반도체 플랫폼 '쿤룬신', 미니맥스 신모델 ‘M3.1 플래시’ 즉시 구동 성공 AI News Korea

Story thread

1. 2026-10-19 MiniMax reportedly releases 'M3.1 Flash', with model weights due October 23 (community report)
2. 2026-10-19 MiniMax launches M3.1 Flash API, a 428B-total/23B-active MoE that ran early on OpenRouter as 'Space Bunny' (community reports)
3. 2026-10-23 MiniMax M3.1 Flash follow-up: weights released on Hugging Face as promised (community report)
4. 2026-10-25 MiniMax M3.1 Flash follow-up: Baidu-affiliated chipmaker Kunlunxin says the model runs on its accelerators, as Chinese chips race for Day 0 support (report)