Mistral Large 4 Preview vs Qwen3.8 27B Xhigh compared

This comparison puts Mistral's Mistral Large 4 Preview next to Alibaba's Qwen3.8 27B (Xhigh). Mistral Large 4 Preview streams output at about 130 tokens per second, 3.4 times the rate of Qwen3.8 27B Xhigh. What follows is built entirely from tracked data, so where the catalogue is silent, this page says so instead of guessing.

Throughput and time to first token

Measured output speed is 130 tokens per second for Mistral Large 4 Preview and 38 tokens per second for Qwen3.8 27B Xhigh. On long outputs, that is a real advantage for Mistral Large 4 Preview. Time to first token is 1.9 seconds for Mistral Large 4 Preview and 4.1 seconds for Qwen3.8 27B Xhigh. Qwen3.8 27B Xhigh waits 2.1 times as long before its first token.

Put together, a 1,000-token answer would take about 9.6 seconds from Mistral Large 4 Preview and 30 seconds from Qwen3.8 27B Xhigh, counting the first-token wait.

When they came out

Qwen3.8 27B Xhigh came first, on August 14, 2026; Mistral Large 4 Preview followed on October 6, 2026.

Pricing depends on your mix

The list prices are $0.70 in and $2.10 out for Mistral Large 4 Preview, and $0.50 in and $3.00 out for Qwen3.8 27B Xhigh, per million tokens. The rates cross over. Qwen3.8 27B Xhigh is cheaper to feed and Mistral Large 4 Preview is cheaper to hear back from.

A sample workload of 1 million input tokens and 200,000 output tokens costs about the same on either: $1.12 against $1.10. For classifying 100,000 short messages at 150 input tokens and 5 output tokens each, Qwen3.8 27B Xhigh would come in 22% lower: $11.55 for Mistral Large 4 Preview against $9.00 for Qwen3.8 27B Xhigh.

Reasoning and overall scores

Mistral Large 4 Preview records 25.3 on the intelligence index and Qwen3.8 27B Xhigh records 20.1. That is a clear margin for an index that averages many evaluations.

Across the 109 active models with an intelligence score in the catalogue, Mistral Large 4 Preview ranks 67th and Qwen3.8 27B Xhigh ranks 78th.

There are no coding or math index scores for either model in the catalogue, which keeps this comparison at the aggregate level.

Score per dollar

A blunt value measure, blended price over intelligence score, puts Mistral Large 4 Preview at $0.042 and Qwen3.8 27B Xhigh at $0.056 per point. Mistral Large 4 Preview scores higher and costs less per point, so on price per point there is nothing to trade off.