Kimi K3.1 Pro
Terminal-Bench 4.0: 65.7%, against GPT-6 Astra 58.2% and Claude Opus 5.5 66.4%.
65.7% vs 66.4%Terminal-Bench 4.0, K3.1 Pro vs Claude Opus 5.5 (GPT-6 Astra 58.2%), per Moonshot
Open weights
Week 43 · 19-25 Oct
2 sources filed · EN 2
B B Raon-1-mini-preview (Hugging Face model card)
Architecture: 128 routed experts, 6 active plus 1 shared; Raon Delta Attention modifies Kimi Delta Attention; Raon-mHC cuts the cost of DeepSeek's manifold-constrained hyper-connections; no positional encoding; 259,840-token vocabulary; about 8 KiB of KV cache per token in BF16.
28.8B / 4.3Btotal versus active parameters per token in the hybrid mixture-of-experts model
Week 42 · 12-18 Oct
1 source filed · ZH 1
C Zhipu's GLM-Legal tops legal benchmarks
Training: mid-training on legal corpora on top of GLM-5.3-Flash, then reinforcement learning across tens of thousands of legal task environments, with the agent framework iterated alongside.
19.2%Harvey LAB tasks where GLM-Legal met every scoring criterion (top open-weight model on Vals AI)
Week 41 · 5-11 Oct
2 sources filed · EN 2
C Introducing Mistral Large 4 (public preview)
1 trillion parameters, 52 billion active; trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs; training data spans more than 160 languages.
1 trillion / 52 billiontotal and active parameters of Mistral Large 4
C Reflection AI announces Beam (501B/23B active) open-weight reasoning model
Architecture: sparse mixture-of-experts, 501B total parameters, 23B active per inference, text only.
501B / 23Btotal and active parameters per inference in the sparse mixture-of-experts model