Qwen3.8 在 GPT-5.5 Pro 推理 prefill 实验中表现出与 GPT 系模型的显著重合

Reasoning prefills on a few open models, v1.1

A follow-up to Reasoning prefills on a few open models and Stolen Thoughts

This v1.1 reruns the reasoning-prefill experiment with GPT-5.5 Pro as the teacher.

For each problem, I generated two responses from each target model:

an ordinary, unprefilled response; and
a response starting with the first 1% of GPT-5.5 Pro's reasoning, inserted into the target model's reasoning channel.

The visible answer remained freely generated. I then measured how much of the teacher's visible answer appeared in the first 100 tokens of the target model's answer. As in the previous post, each score is the mean of unigram, bigram, and trigram source recall. Deltas are absolute percentage-point changes.

All problems

The evaluation contains 45 problems: 15 STEM, 15 non-STEM, and 15 synthetic puzzles.

Model	n	Unprefilled	GPT-5.5 Pro reasoning prefill	Delta
DeepSeek V4 Flash	45	27.30%	26.13%	−1.17 pp
Inkling	45	19.99%	20.45%	+0.46 pp
Kimi K3	45	31.11%	35.65%	+4.54 pp
Qwen3.8 A95B	45	16.79%	34.97%	+18.18 pp

Qwen by category

Category	n	Unprefilled	GPT-5.5 Pro reasoning prefill	Delta
STEM	15	19.26%	46.24%	+26.99 pp
Non-STEM	15	20.62%	33.42%	+12.80 pp
Puzzle	15	10.49%	25.23%	+14.75 pp
All	45	16.79%	34.97%	+18.18 pp

Discussion

Qwen barely moved toward Opus 4.8 in the earlier experiment, but moved by +18.18 points toward GPT-5.5 Pro here, including a large effect on the private synthetic puzzles. The data suggest that Qwen may have learned from GPT-5.5 Pro, or from a closely related GPT model, rather than from Opus.

Kimi K3 has the highest overlap with GPT-5.5 Pro both without and with the prefill (31.11% and 35.65%), although the prefill adds only +4.54 points.