$Alibaba(BABA.US)This is the reason I bought Alibaba

Repost:

Benchmark test results for the Qwen3.8-Max preview version have leaked.

According to benchmark test results obtained internally from @Alibaba_Qwen, the Qwen3.8-Max preview version trails Fable 5 by only 12.6 points in coding, while its performance has already surpassed Opus 4.8 Max in actual Cowork agent tasks.

The benchmark covers 400 real-world agent tasks.

Compared to Fable 5, 73.2% of coding task results are identical.

In Cowork tasks, Qwen3.8-Max leads:

- Opus 4.8 Max: 5 points

- Kimi K3: 8 points

- GLM-5.2: 17.1 points

- Previous generation Qwen3.7-Max: 44.4 points

And this is just the preview version.

Sources say that some ready improvements have not yet been deployed, and the final version will be even more powerful.

Qwen3.8's current goal is Fable 5, and Opus has already been surpassed