Unverified50% confidenceFactExact time
SWE-2 is post-trained on Kimi K3 and uses reinforcement learning (RL), optimizing along both the cost and capability dimensions simultaneously.
1
Sources
50%
Confidence
Long-term
Relevance
9/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMoonshot AI stated that Kimi K3's framework and data training methods improved overall scaling efficiency by approximately 2.5 times compared to the previous generation Kimi K2.72% similarUnverifiedKimi K3's level of intelligence was not achieved by piling on compute, but through more sophisticated architecture and training strategies.71% similarUnverifiedHY3 introduces improved anti-hallucination techniques, more reliable tool-calling capabilities, and employs higher-quality post-training data and an enhanced reinforcement learning strategy.68% similarUnverifiedKimi K2.6能根据问题复杂度自适应调整输出长度，这一能力通过Post Training阶段的长度校准技术实现67% similarUnverified后训练优化包括指令微调、RLHF或DPO策略调整、系统提示词工程改进及工具调用和多模态融合能力强化66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/918904API
curl https://kongchang.com/api/v1/knowledge/claims/918904MCP
get_claim(id=918904)