Open-Weight LLM Models — Live Ranking, Benchmarks & Specs
Explore, filter and compare open-weight language models with live specs, benchmarks and license data
Nanbeige4.3-33B
Nanbeige
dense Llama-style Transformer decoder with GQA
Nanbeige4.3
Apache License 2.0
Openness Score
1.0
Spark-X2.6-4B
SparkLLM (XHToken)
Hybrid attention transformer: 1 full-attention layer combined with 3 sliding-window attention (SWA) layers, natively supporting up to 1M-token context
Spark-X2.6
Apache License 2.0
Openness Score
1.0
Flint-40B-A2.5B
Recursive
Decoder-only mixture-of-experts Transformer (hybrid sliding-window + full attention)
Flint
Apache License 2.0
Openness Score
1.0
Raon-1 mini
KRAFTON AI
hybrid Mixture-of-Experts (MoE) Transformer decoder with Raon Delta Attention (18 linear attention layers) and Gated MLA (7 full-attention layers), Raon-mHC residual connections, NoPE
Raon
Apache License 2.0
Openness Score
1.0
Muse-Spark-Open-816B-A42B
Meta Superintelligence Lab
Mixture-of-Experts Transformer with Perception Encoder
Muse Spark
Muse Model License
Openness Score
0.7
Open Weights
Mamba2-Transformer Hybrid Latent Mixture of Experts (LatentMoE) with C-RADIO v4-H vision encoder
Nemotron
OpenMDW License Agreement v1.1
Openness Score
0.7
Kolibri-1
Aleph Alpha
Sparse MoE reasoning model with hybrid 4:1 sliding-window/full GQA attention
Kolibri
Apache License 2.0
Openness Score
1.0
Clef
Cloudflare
Qwen3.8-27B multimodal backbone with joint schema head for typed decision outputs
Qwen3.8
Apache License 2.0
Openness Score
1.0
Clef-Flash
Cloudflare
multimodal hybrid linear attention transformer + joint schema head
Qwen3.5
Apache License 2.0
Openness Score
1.0
Bonsai 2 27B — GGUF
Prism ML
Hybrid attention (~75% linear / ~25% full attention), SwiGLU MLP, RoPE, RMSNorm
Bonsai
Apache License 2.0
Openness Score
1.0
Xing4.0-29B-A4B
China Telecom AI
Sparse MoE Transformer with mHC + MLA + MTP
Xing4.0
Apache License 2.0
Openness Score
1.0
DeepSeek-V4-Flash-Vision-Exp
DeepSeek
MoE with vision encoder and aligner, DFlash attention, DSpark speculative decoding
DeepSeek
MIT License
Openness Score
1.0
Spark-X2.5-4B
SparkLLM (XHToken)
Hybrid attention transformer: 1 full-attention layer combined with 3 sliding-window attention (SWA) layers, natively supporting up to 1M-token context
Spark-X2.5
Apache License 2.0
Openness Score
1.0
Spark-X2.5-1.7B
SparkLLM (XHToken)
Hybrid attention transformer: 1 full-attention layer combined with 3 sliding-window attention (SWA) layers, natively supporting up to 1M-token context
Spark-X2.5
Apache License 2.0
Openness Score
1.0
Hy4 preview
Tencent
MoE (78 layers) with Gated DeepSeek Sparse Attention (Gated DSA) + IndexCache cross-layer sparse index reuse, iHC (identity Hyper-Connections) residual streams, 1 native MTP layer (10B total / 0.7B active) for speculative decoding
Hunyuan
Apache License 2.0
Openness Score
1.0
GLM-5.3-Flash
Zhipu AI
Hybrid sparse + linear attention MoE with mHC
GLM
MIT License
Openness Score
1.0