DeepSeek-V3.2
DeepSeekBy implementing a robust RL protocol and scaling post-training compute, DeepSeek-V3.2 performs comparably to GPT-5.
1 more from this model
A scalable reinforcement learning post-training framework further improves reasoning
general family · filed under post-training
A broadly described framework that scales reinforcement-learning post-training to improve model performance, without a more specific mechanism established by the evidence.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See post-training for the methods filed beside it.
Also called Scalable Reinforcement Learning Framework.
2 quotes from 2 sources, one card per model, each opening with its strongest.
By implementing a robust RL protocol and scaling post-training compute, DeepSeek-V3.2 performs comparably to GPT-5.
A scalable reinforcement learning post-training framework further improves reasoning