Reinforcement Learning Interpretability Qwen

Qwen-Scope

A Qwen language model release.

Visit site

A language model release in the Qwen line from Alibaba Cloud.

Pricing

Model profile · Current snapshot

Input / 1M

$0.30

Output / 1M

$1.90

Blended / 1M

$0.70

Output TPS

50.21

TTFT (s)

1.07

Model Intelligence

Arena ranking

Benchmarkable

Yes

Model level

release

Intelligence Index

20.8

Coding Index

16.5

Math Index

70.7

MMLU Pro

0.82

GPQA

0.71

HLE

0.06

LiveCodeBench

0.59

SciCode

0.36

AIME 2025

0.71

IFBench

0.43

LCR

0.32

TerminalBench Hard

0.07

TAU2

0.35

Recent stories

1 linked story

releasePRIMARY2026-04-30

Qwen-Scope releases SAE toolkit for Qwen3.5-27B steering

Alibaba’s Qwen team released Qwen-Scope, an open sparse-autoencoder suite for Qwen3.5-27B that can steer outputs, surface repetition features, and compare benchmark feature overlap. The toolkit turns interpretability artifacts into debugging, data-generation, and evaluation workflows.