Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-08-07 · high
It lifts Qwen3 8B non-thinking accuracy 10.7 points by building pseudo-answers via majority vote, then distilling them back into itself.
Why now: It posted to arXiv in August 2026, claiming to remove the last external dependency left in on-policy self-distillation training.
U-OPSD trains reasoning models to solve math with no answer key, teacher model, or reward feedback. The method comes from a paper posted to arXiv in August 2026.
That removes one of the standard costs of training reasoning models: labeled answers or a separate reward model to check against. It's aimed at math benchmarks with one checkable right answer, per the paper.
GRPO is the reinforcement learning method this kind of self-distillation research gets measured against. In thinking mode, U-OPSD beat it by 0.7 to 1.1 points across the same benchmark set, without GRPO's environment feedback.
The method samples multiple rollouts per problem, then builds a pseudo-solution by majority vote under a self-consistency threshold. It conditions a teacher distribution on the shortest pseudo-solution, then distills that into prefixes of the model's longest incorrect completion. The label is the model's own consensus answer, not a verified one.
Across AIME24, AIME25, HMMT25, MATH500 and AMC23, U-OPSD lifted Qwen3 non-thinking mode accuracy 8.5 points at 4B and 10.7 points at 8B. That beat supervised OPSD, per the paper.
Every one of those benchmarks has a single, checkable right answer. The paper doesn't say whether majority-vote consensus holds up on reasoning tasks without one clean answer to converge on. That's most of what reasoning models get used for outside math contests.
Each link below shares sources, entities, or timing with this story.
Unsloth released GRPO / Shared entities / Earlier coverage / Tension
Linked by a graph relationship (Unsloth released GRPO); both cover GRPO, Qwen3; earlier GRPO coverage from 2026-06-20.
Unsloth released GRPO / Shared entity: Qwen3 / Earlier coverage / Tension
Linked by a graph relationship (Unsloth released GRPO); both cover Qwen3; earlier Qwen3 coverage from 2026-04-23.
Unsloth released GRPO / Shared entity: Qwen3 / Earlier coverage
Linked by a graph relationship (Unsloth released GRPO); both cover Qwen3; earlier Qwen3 coverage from 2026-08-03.
Linked by a graph relationship (Unsloth released GRPO); both cover Qwen3; earlier Qwen3 coverage from 2026-07-11.
Unsloth released GRPO
Linked by a graph relationship (Unsloth released GRPO).
Shared entities / Same source domain / Earlier coverage / Tension
Both cover Qwen3, Self; reported by the same outlet (arxiv.org); earlier Qwen3 coverage from 2026-04-21.
GRPO competes with PPO / Shared entity: GRPO / Same source domain / Earlier coverage
Linked by a graph relationship (GRPO competes with PPO); both cover GRPO; reported by the same outlet (arxiv.org).
Shared entities / Shared topic / Earlier coverage
Both cover AIME25, Qwen3; overlapping topics (aime25, beating); earlier AIME25 coverage from 2026-08-06.