MiniMax-M3 vs Qwen3.8 27B Xhigh: which is faster?

MiniMax-M3 from MiniMax and Qwen3.8 27B (Xhigh) from Alibaba land within 1.2 points of each other on the intelligence index. The biggest gap is output speed, where MiniMax-M3 records 121 tokens per second and Qwen3.8 27B Xhigh 38 tokens per second. What follows is built entirely from tracked data, so where the catalogue is silent, this page says so instead of guessing.

How fast they respond

Once a response is under way, MiniMax-M3 produces roughly 121 tokens per second against 38 tokens per second for Qwen3.8 27B Xhigh. MiniMax-M3 is ahead by 214% on throughput. Before any text appears, MiniMax-M3 takes 0.82 seconds and Qwen3.8 27B Xhigh 4.1 seconds. MiniMax-M3 gets to its first token 5 times as fast.

Add the two numbers up for a 1,000-token reply and MiniMax-M3 finishes first: 9.1 seconds for MiniMax-M3, 30 seconds for Qwen3.8 27B Xhigh.

Benchmark scores

MiniMax-M3 records 18.9 on the intelligence index and Qwen3.8 27B Xhigh records 20.1. It is a small margin for an index that averages many evaluations. Put another way, MiniMax-M3 would need 6% more on this index to draw level.

Among 109 active scored models, MiniMax-M3 sits 79th and Qwen3.8 27B Xhigh 78th.

Neither model has a separate coding or math index recorded yet, so those narrower comparisons will have to wait for more data.

Release timing

Qwen3.8 27B Xhigh arrived on August 14, 2026, about 2 months after MiniMax-M3 (June 1, 2026).

What they cost

The list prices are $0.30 in and $1.20 out for MiniMax-M3, and $0.50 in and $3.00 out for Qwen3.8 27B Xhigh, per million tokens. That makes MiniMax-M3's output 60% cheaper.

A reference job of 1 million input tokens and 200,000 output tokens costs $0.54 with MiniMax-M3 and $1.10 with Qwen3.8 27B Xhigh, so MiniMax-M3 saves $0.56. For a support assistant answering 10,000 conversations of 2,000 input and 500 output tokens each, MiniMax-M3 would come in 52% lower: $12.00 for MiniMax-M3 against $25.00 for Qwen3.8 27B Xhigh. On a 3:1 blend of input to output tokens, that works out to $0.525 versus $1.125 per million.

Is the price in proportion?

Dividing the blended price by the intelligence index gives a rough cost per point: $0.028 for MiniMax-M3 and $0.056 for Qwen3.8 27B Xhigh, per million tokens. So Qwen3.8 27B Xhigh buys its higher score at a premium, while MiniMax-M3 is the value pick, paying 50% less per point.