Kimi K3.1 disappoints: analysts say Moonshot trails DeepSeek on RL progress
scaling01 · x · 2026-10-25
- scaling01 amplifies tphuang's take that many find Kimi K3.1 disappointing.
- K3.1 is judged the best released Chinese model, but progress over K3.0 in three months is underwhelming — Opus 5.0/GPT-6 Sol level rather than Opus 5.5/6.1 level.
- On cyber and some SWE tasks, DeepSeek's 1017 is actually better; DeepSeek has progressed faster with RL, and its next flash update should catch up on SWE and work tasks.
- tphuang speculates Moonshot is pouring energy into a larger K4: Moonshot is best at pretraining, worst at post-training among top Chinese labs, so its best bet is keep scaling up.
More from Models
- Users mourn the lost 'rhythm' of early Opus 4.5 checkpoints — repligate · 2026-10-25
- Reflection AI's open-weights Beam served 10 trillion tokens in one week — zacharynado · 2026-10-25
- Dimitris Papail: Raion 'incredibly strong for its size' as from-scratch pretraining spreads — DimitrisPapail · 2026-10-25
- Eric Jang: abliterated models show 'astounding' cyber capabilities — ericjang11 · 2026-10-25
- ChatGPT unexpectedly outputs middle-finger emoji saying it 'fails to process' the message — AYSomkiv · 2026-10-25
- Community org claims to run "DeepSeek v4.1 flash" locally via pruning — unverified — amplifiedamp · 2026-10-25