Hacker News front pageSep 22, 2026, 12:02 UTC+8

Xiaomi's MiMo-V2.6-Pro tops AA Intelligence Index, fast but verbose

MiMo-v2.6-Pro: Intelligence, Performance and Price Analysis

Artificial Analysis ranks Xiaomi's MiMo-V2.6-Pro #1 out of 114 models with a score of 46. It's a 1T total / 42B active parameter open-weight model with text, image, speech, and video input. Output speed is 125 tokens/sec, but it's verbose—generating 140M tokens during evaluation. Pricing: $0.43/M input, $0.87/M output; the full eval cost $206.66.

Why it matters: Xiaomi's MiMo-v2.6-Pro hits #1 on Artificial Analysis' intelligence index with 1T params, 42B active, 125 tok/s, and $0.43/M input. It's the first Chinese open-weight model to top a major independent benchmark, making it a strong reference for model selection. Score stays at 8...

Read the original ↗Export Markdown

Take

The reason to click: an open-weight model from Xiaomi just took the #1 spot on Artificial Analysis's overall leaderboard, scoring 46 on their intelligence index and beating a bunch of proprietary models. 1 trillion total parameters with only 42B active per token—same design philosophy as DeepSeek V3, where a massive base model only runs a fraction during inference to save compute.

Speed is genuinely good at 125 tokens/sec, but it generated 140M tokens during evaluation, so it's on the chatty side. Pricing at $0.43/$0.87 per million tokens is pricier than DeepSeek V3, but with an MIT license and the ability to run locally, the premium makes sense.

I'd discount the ranking a bit: AA's eval suite weights reasoning and coding heavily, which may not reflect everyday chat quality. The post also doesn't break out performance on speech and video inputs, so hold the excitement there. Overall, another strong contender from the open-weight camp—just be clear about what "#1" is measuring.

Source: Hacker News front page · artificialanalysis.ai
Published: Sep 22, 2026, 12:02 UTC+8 · 8 days ago