https://llm-atlas-7or.pages.dev Browse, filter and compare 100+ open-weight LLM architectures: diagrams, attention mechanisms, MoE layouts, KV-cache sizes and fact sheets, sourced from Sebastian Raschka's LLM Architecture Gallery. Wed, 04 Nov 2026 16:15:11 GMT https://llm-atlas-7or.pages.dev/models/mistral-large-4/ https://llm-atlas-7or.pages.dev/models/mistral-large-4/ Thu, 29 Oct 2026 00:00:00 GMT Mistral AI · Sparse MoE · 1T total, 52B active (5.2% active) · 1M context. Mistral's trillion-parameter multimodal MoE combines MLA with sparse token selection and cross-layer KV sharing. https://llm-atlas-7or.pages.dev/models/lfm3-1-7b/ https://llm-atlas-7or.pages.dev/models/lfm3-1-7b/ Tue, 27 Oct 2026 00:00:00 GMT Liquid AI · Dense hybrid · 1.73B parameters · 256K context. Liquid AI's 1.73B dense hybrid combines short convolutions with sliding-window and global GQA in a 24-layer stack. https://llm-atlas-7or.pages.dev/models/minimax-m3-1-flash-428b/ https://llm-atlas-7or.pages.dev/models/minimax-m3-1-flash-428b/ Mon, 19 Oct 2026 00:00:00 GMT MiniMax · Sparse MoE · 428B total, 23B active (5.4% active) · 1M context. MiniMax M3 successor with sparse attention in all 60 decoder layers, quantization-aware training, and a DSpark draft head for speculative decoding. https://llm-atlas-7or.pages.dev/models/recursive-flint-40b-a2-5b/ https://llm-atlas-7or.pages.dev/models/recursive-flint-40b-a2-5b/ Sun, 18 Oct 2026 00:00:00 GMT Recursive · Sparse MoE · 39.8B total, 2.5B active (6.3% active) · 256K context. Recursive's coding model uses two dense FFN layers followed by 47 MoE layers. The active-parameter count includes the output head. https://llm-atlas-7or.pages.dev/models/reflection-beam-501b/ https://llm-atlas-7or.pages.dev/models/reflection-beam-501b/ Fri, 16 Oct 2026 00:00:00 GMT Reflection · Sparse MoE · 501.3B total, 23.4B active (4.7% active) · 1M context. Text-only MoE decoder with three dense FFN layers followed by 49 expert layers. Each MoE layer activates 8 of 256 experts without a shared expert. https://llm-atlas-7or.pages.dev/models/dory1-0-4b/ https://llm-atlas-7or.pages.dev/models/dory1-0-4b/ Fri, 16 Oct 2026 00:00:00 GMT NVIDIA · Dense · 4B parameters · 256K context. NVIDIA's dense looped transformer applies a shared 20-block core twice between 8-block input and output stacks, using normalized GPT (nGPT) residual updates. https://llm-atlas-7or.pages.dev/models/muse-spark-open-816b-a42b/ https://llm-atlas-7or.pages.dev/models/muse-spark-open-816b-a42b/ Thu, 15 Oct 2026 00:00:00 GMT Meta · Sparse MoE · 816B total, 42B active (5.1% active) · 1M context. Meta's multimodal MoE model combines 62 latent-expert blocks with gated GQA, local-only RoPE, global NoPE, and learned residual gates; the figure focuses on the text decoder. https://llm-atlas-7or.pages.dev/models/nemotron-3-5-super-vl-120b-a12b/ https://llm-atlas-7or.pages.dev/models/nemotron-3-5-super-vl-120b-a12b/ Thu, 15 Oct 2026 00:00:00 GMT NVIDIA · Hybrid MoE · 120B total, 12B active (10% active) · 1M context. NVIDIA's vision-language model extends the Nemotron 3 Super backbone with C-RADIO vision encoding and shared-weight multi-token prediction. The figure shows the language backbone, with the vision encoder and MTP omitted. https://llm-atlas-7or.pages.dev/models/ling-3-1-flash/ https://llm-atlas-7or.pages.dev/models/ling-3-1-flash/ Tue, 13 Oct 2026 00:00:00 GMT InclusionAI · Sparse hybrid · 560B total, 25B active (4.5% active) · 1M context. Hybrid MoE that extends Ling 3.0 Flash to 72 layers, with seven Kimi Delta Attention layers for every gated MLA layer. https://llm-atlas-7or.pages.dev/models/kolibri-1-78b/ https://llm-atlas-7or.pages.dev/models/kolibri-1-78b/ Sat, 03 Oct 2026 00:00:00 GMT Aleph Alpha · Sparse MoE · 78B total, 3.46B active (4.4% active) · 256K context. Aleph Alpha's German-English reasoning model uses 50 MoE blocks with a 4:1 mix of sliding-window and global GQA. https://llm-atlas-7or.pages.dev/models/naive-n0-5-flash-309b/ https://llm-atlas-7or.pages.dev/models/naive-n0-5-flash-309b/ Sun, 27 Sep 2026 00:00:00 GMT NaiveAI · Sparse MoE · 309B total, 15.5B active (5.0% active) · 1M context. MiMo-V2.5-derived sparse MoE that replaces its global attention layers with GQA-based DeepSeek Sparse Attention for a native 1M-token context. https://llm-atlas-7or.pages.dev/models/ember-1/ https://llm-atlas-7or.pages.dev/models/ember-1/ Wed, 23 Sep 2026 00:00:00 GMT Fireworks AI · Sparse hybrid · 2.8T total, 104B active (3.7% active) · 1M context. Fireworks' Kimi K3 derivative, post-trained for shorter reasoning traces. Architecture details follow the published Kimi K3 base model; the diagram focuses on its text backbone. https://llm-atlas-7or.pages.dev/models/xiaomi-mimo-v2-6-flash-rl-309b/ https://llm-atlas-7or.pages.dev/models/xiaomi-mimo-v2-6-flash-rl-309b/ Mon, 21 Sep 2026 00:00:00 GMT Xiaomi · Sparse omnimodal MoE · 309B total, 15B active (4.9% active) · 1M context. Omnimodal sparse MoE model with the MiMo-V2.5 text backbone, mixed-domain reinforcement learning, and a five-layer DFlash speculative decoder. https://llm-atlas-7or.pages.dev/models/xiaomi-mimo-v2-6-pro-rl-1-02t/ https://llm-atlas-7or.pages.dev/models/xiaomi-mimo-v2-6-pro-rl-1-02t/ Mon, 21 Sep 2026 00:00:00 GMT Xiaomi · Sparse omnimodal MoE · 1.02T total, 42B active (4.1% active) · 1M context. Omnimodal sparse MoE model with the MiMo-V2.5-Pro text backbone, mixed-domain reinforcement learning, and a five-layer DFlash speculative decoder. https://llm-atlas-7or.pages.dev/models/step-5-preview-600b/ https://llm-atlas-7or.pages.dev/models/step-5-preview-600b/ Sun, 20 Sep 2026 00:00:00 GMT StepFun · Sparse MoE · 600B total, 27B active (4.5% active) · 1M context. StepFun's 92-layer MoE pairs 512-token sliding-window attention with sparse GQA for a 1M-token context window; the figure shows the text decoder. https://llm-atlas-7or.pages.dev/models/deepseek-v4-1-flash/ https://llm-atlas-7or.pages.dev/models/deepseek-v4-1-flash/ Thu, 10 Sep 2026 00:00:00 GMT DeepSeek · Sparse MoE (causal encoder-decoder) · 552B backbone + 196B Engram; 8B active prefill, 16B active decode (2.9% of backbone) · 1M context. Causal encoder-decoder MoE with CSA2 cache sharing, Engram memory, and a million-token context. The figure shows the text backbone. https://llm-atlas-7or.pages.dev/models/tencent-hy4-preview-770b-a49b/ https://llm-atlas-7or.pages.dev/models/tencent-hy4-preview-770b-a49b/ Fri, 28 Aug 2026 00:00:00 GMT Tencent · Sparse MoE · 770B backbone, 49B active (6.4% active); plus 10B MTP · 1M context. Tencent's 770B sparse MoE combines gated DeepSeek Sparse Attention, cross-layer index reuse, and four-stream identity Hyper-Connections. https://llm-atlas-7or.pages.dev/models/glm-5-3-flash/ https://llm-atlas-7or.pages.dev/models/glm-5-3-flash/ Wed, 26 Aug 2026 00:00:00 GMT Z.ai · Sparse hybrid · 320B total, 18B active (5.6% active) · 1M context. Natively multimodal GLM refresh, previously previewed as Ox Alpha, that replaces most MLA/DSA layers with Kimi Delta Attention and adds four-stream mHC. https://llm-atlas-7or.pages.dev/models/qwen3-8-flash-next-125b-a6b/ https://llm-atlas-7or.pages.dev/models/qwen3-8-flash-next-125b-a6b/ Wed, 26 Aug 2026 00:00:00 GMT Alibaba Cloud · Sparse hybrid · 125B core, 6B active (4.8% active); plus 51B n-gram embeddings and 4B MTP · 256K context. Experimental vision-language MoE whose text backbone combines Gated DeltaNet, micro-block QSA, four-branch gated residuals, and n-gram embeddings; the figure omits the vision encoder and MTP head. https://llm-atlas-7or.pages.dev/models/ornith-1-5-35b-a3b/ https://llm-atlas-7or.pages.dev/models/ornith-1-5-35b-a3b/ Sun, 23 Aug 2026 00:00:00 GMT Ornith · Sparse hybrid · 35B total, 3B active (8.6% active) · 256K context. Uses the same sparse-hybrid decoder geometry as Qwen3.6 35B-A3B, paired with Ornith's self-improvement training recipe. https://llm-atlas-7or.pages.dev/models/qwen3-8-27b/ https://llm-atlas-7or.pages.dev/models/qwen3-8-27b/ Fri, 14 Aug 2026 00:00:00 GMT Alibaba Cloud · Dense hybrid · 27B parameters · 256K context. Dense 27B native vision-language model with a 64-layer Gated DeltaNet and Gated Attention language backbone; the figure omits the vision encoder and MTP head. https://llm-atlas-7or.pages.dev/models/nemotron-3-5-lightning-30b-a3b/ https://llm-atlas-7or.pages.dev/models/nemotron-3-5-lightning-30b-a3b/ Tue, 11 Aug 2026 00:00:00 GMT NVIDIA · Hybrid MoE · 30B total, 3B active (10% active) · 1M context. NVIDIA's 30B hybrid model keeps 3B parameters active and extends the Nemotron 3 Nano backbone with a 1M-token context window and multi-token prediction. https://llm-atlas-7or.pages.dev/models/muse-glimmer-30b/ https://llm-atlas-7or.pages.dev/models/muse-glimmer-30b/ Mon, 10 Aug 2026 00:00:00 GMT Meta · Dense · 29.6B parameters (including perception encoder) · 128K context. Meta's dense 29.6B multimodal model combines a 52-layer sandwich-norm decoder with gated GQA and a separate perception encoder; the figure focuses on the text backbone. https://llm-atlas-7or.pages.dev/models/ling-3-0-flash/ https://llm-atlas-7or.pages.dev/models/ling-3-0-flash/ Tue, 04 Aug 2026 00:00:00 GMT InclusionAI · Sparse hybrid · 124B total, 5.1B active (4.1% active) · 256K context. Hybrid MoE that alternates five Kimi Delta Attention layers with one gated MLA layer. https://llm-atlas-7or.pages.dev/models/kimi-k3/ https://llm-atlas-7or.pages.dev/models/kimi-k3/ Mon, 27 Jul 2026 00:00:00 GMT Moonshot AI · Sparse hybrid · 2.8T total, 104B active (3.7% active) · 1M context. 2.8T sparse hybrid that combines a 3:1 KDA and Gated MLA stack with Block Attention Residuals and Stable LatentMoE. It is also native multimodal, but the figure focuses on the text portion. https://llm-atlas-7or.pages.dev/models/solar-open-2/ https://llm-atlas-7or.pages.dev/models/solar-open-2/ Wed, 22 Jul 2026 00:00:00 GMT Upstage · Sparse hybrid · 250B total, 15B active (6% active) · 1M context. Upstage's sparse hybrid alternates one gated GQA layer with three Kimi Delta Attention layers and reaches a 1M-token context without positional embeddings. https://llm-atlas-7or.pages.dev/models/antares-1b/ https://llm-atlas-7or.pages.dev/models/antares-1b/ Tue, 21 Jul 2026 00:00:00 GMT Cisco · Dense · 1B parameters · 128K context. Cisco Foundation AI's compact terminal agent uses a dense Granite 4.0 backbone and is post-trained with SFT and GRPO for vulnerability localization. https://llm-atlas-7or.pages.dev/models/laguna-s-2-1/ https://llm-atlas-7or.pages.dev/models/laguna-s-2-1/ Tue, 21 Jul 2026 00:00:00 GMT Poolside · Sparse MoE · 118B total, 8B active (6.8% active) · 1M context. Poolside's 118B sparse MoE scales the Laguna XS recipe to 48 layers and 1M context while activating about 8B parameters per token. https://llm-atlas-7or.pages.dev/models/motif-3-beta/ https://llm-atlas-7or.pages.dev/models/motif-3-beta/ Tue, 21 Jul 2026 00:00:00 GMT Motif Technologies · Sparse MoE · 314B total, 13B active (4.1% active) · 256K context. Large sparse MoE that combines Grouped Differential Latent Attention, per-expert PolyNorm, and modified mHC in a 53-layer decoder. https://llm-atlas-7or.pages.dev/models/nanbeige-4-2-3b/ https://llm-atlas-7or.pages.dev/models/nanbeige-4-2-3b/ Tue, 21 Jul 2026 00:00:00 GMT BOSS Zhipin · Dense · 4B total, 3B non-embedding · 256K context. Compact agentic model that runs the same 22-layer dense transformer stack twice, resulting in 44 transformer blocks (layers) without duplicating the layer weights. https://llm-atlas-7or.pages.dev/models/btl-3-27b/ https://llm-atlas-7or.pages.dev/models/btl-3-27b/ Mon, 20 Jul 2026 00:00:00 GMT Bad Theory Labs · Dense hybrid · 27B base + 934 MB LoRA adapter · 256K context. Agent-oriented Qwen3.6-27B derivative that adds a rank-32 LoRA adapter for coding, structured tool use, and long multi-turn execution. https://llm-atlas-7or.pages.dev/models/inkling/ https://llm-atlas-7or.pages.dev/models/inkling/ Wed, 15 Jul 2026 00:00:00 GMT Thinking Machines Lab · Sparse MoE · 975B total, 41B active (4.2% active) · 1M context. Thinking Machines Lab's native multimodal model combines a 66-layer sparse MoE decoder with a 5:1 mix of local and global GQA. https://llm-atlas-7or.pages.dev/models/soofi-s-30b-a3b/ https://llm-atlas-7or.pages.dev/models/soofi-s-30b-a3b/ Fri, 10 Jul 2026 00:00:00 GMT SOOFI Consortium · Hybrid MoE · 31.6B total, 3.2B active (10.1% active) · 1M context. German-English base model that retrains the Nemotron 3 Nano architecture on 26.68T tokens with a fully documented data mixture. https://llm-atlas-7or.pages.dev/models/laguna-xs-2-1/ https://llm-atlas-7or.pages.dev/models/laguna-xs-2-1/ Thu, 02 Jul 2026 00:00:00 GMT Poolside · Sparse MoE · 33B total, 3B active (9.1% active) · 256K context. Poolside's 33B Laguna refresh extends the sparse MoE architecture to 262k context while keeping about 3B parameters active per token. https://llm-atlas-7or.pages.dev/models/glm-5-2/ https://llm-atlas-7or.pages.dev/models/glm-5-2/ Wed, 17 Jun 2026 00:00:00 GMT Z.ai · Sparse MoE · 744B total, 40B active (5.4% active) · 1M context. Long-context GLM refresh that keeps the 744B sparse MoE backbone and adds IndexShare for 1M-token DSA inference. https://llm-atlas-7or.pages.dev/models/vibethinker-3b/ https://llm-atlas-7or.pages.dev/models/vibethinker-3b/ Mon, 15 Jun 2026 00:00:00 GMT Weibo · Dense · 3B parameters · 128K context. Qwen2.5-Coder-based dense reasoner whose gains mainly come from verifier-driven post-training rather than a new block design. https://llm-atlas-7or.pages.dev/models/minimax-m3-428b/ https://llm-atlas-7or.pages.dev/models/minimax-m3-428b/ Sat, 13 Jun 2026 00:00:00 GMT MiniMax · Sparse MoE · 428B total, 23B active (5.4% active) · 1M context. Native multimodal MiniMax model that adds sparse attention to an M2-style MoE backbone for 1M-token context. https://llm-atlas-7or.pages.dev/models/kimi-k2-7-code/ https://llm-atlas-7or.pages.dev/models/kimi-k2-7-code/ Fri, 12 Jun 2026 00:00:00 GMT Moonshot AI · Sparse MoE · 1T total, 32B active (3.2% active) · 256K context. Coding-focused K2.6 successor that keeps the same 1T sparse MoE text backbone and emphasizes long-horizon software engineering agents. https://llm-atlas-7or.pages.dev/models/north-mini-code-30b-a3b/ https://llm-atlas-7or.pages.dev/models/north-mini-code-30b-a3b/ Fri, 05 Jun 2026 00:00:00 GMT Cohere · Sparse MoE · 30B total, 3B active (10% active) · 256K context. Cohere's compact agentic coding MoE adapts the parallel Cohere2-MoE block to a 30B total, 3B active model. https://llm-atlas-7or.pages.dev/models/nemotron-3-ultra-550b-a55b/ https://llm-atlas-7or.pages.dev/models/nemotron-3-ultra-550b-a55b/ Thu, 04 Jun 2026 00:00:00 GMT NVIDIA · Hybrid MoE · 550B total, 55B active (10% active) · 1M context. NVIDIA's Ultra model scales the Nemotron 3 hybrid stack to 550B total parameters while keeping 55B active per token. https://llm-atlas-7or.pages.dev/models/gemma-4-12b/ https://llm-atlas-7or.pages.dev/models/gemma-4-12b/ Wed, 03 Jun 2026 00:00:00 GMT Google · Dense · 12B parameters · 128K context. Dense Gemma 4 12B Unified uses an encoder-free multimodal input path, while its text decoder keeps the 31B-style local-global attention recipe. https://llm-atlas-7or.pages.dev/models/jetbrains-mellum2-thinking-12b-a2-5b/ https://llm-atlas-7or.pages.dev/models/jetbrains-mellum2-thinking-12b-a2-5b/ Mon, 01 Jun 2026 00:00:00 GMT JetBrains · Sparse MoE · 12B total, 2.5B active (20.8% active) · 128K context. JetBrains' Mellum2 Thinking is a compact reasoning-tuned sparse MoE for code and agentic workflows. https://llm-atlas-7or.pages.dev/models/lfm2-5-1-2b/ https://llm-atlas-7or.pages.dev/models/lfm2-5-1-2b/ Thu, 28 May 2026 00:00:00 GMT Liquid AI · Dense hybrid · 1.2B parameters · 128K context. Liquid AI's 1.2B LFM2.5 model keeps the 16-layer hybrid LIV convolution and GQA stack at a wider hidden size. https://llm-atlas-7or.pages.dev/models/lfm2-5-350m/ https://llm-atlas-7or.pages.dev/models/lfm2-5-350m/ Thu, 28 May 2026 00:00:00 GMT Liquid AI · Dense hybrid · 350M parameters · 128K context. Liquid AI's compact LFM2.5 model combines double-gated LIV convolution blocks with periodic GQA layers. https://llm-atlas-7or.pages.dev/models/lfm2-5-8b-a1b/ https://llm-atlas-7or.pages.dev/models/lfm2-5-8b-a1b/ Thu, 28 May 2026 00:00:00 GMT Liquid AI · Sparse hybrid · 8.3B total, 1.5B active (18.1% active) · 128K context. Liquid AI's sparse LFM2.5 model combines LIV convolution, GQA, and MoE routing with 4 of 32 experts active per token. https://llm-atlas-7or.pages.dev/models/command-a-218b-a25b/ https://llm-atlas-7or.pages.dev/models/command-a-218b-a25b/ Wed, 20 May 2026 00:00:00 GMT Cohere · Sparse MoE · 218B total, 25B active/token (11.5% active) · 128K context. Cohere's open sparse MoE model scales the Command A parallel transformer block to 218B total parameters while keeping 25B active per token. https://llm-atlas-7or.pages.dev/models/zaya1-8b/ https://llm-atlas-7or.pages.dev/models/zaya1-8b/ Wed, 06 May 2026 00:00:00 GMT Zyphra · Sparse MoE · 8.4B total, 760M active/token (9% active) · 128K context. Zyphra's compact reasoning MoE pairs CCA/GQA attention updates with top-1 routed expert FFN updates to keep fewer than 1B parameters active per token. https://llm-atlas-7or.pages.dev/models/granite-4-1-30b/ https://llm-atlas-7or.pages.dev/models/granite-4-1-30b/ Wed, 29 Apr 2026 00:00:00 GMT IBM · Dense · 30B parameters · 128K context. IBM's dense 30B Granite 4.1 model keeps a straightforward long-context GQA transformer stack and improves tool calling, instruction following, and chat behavior through post-training. https://llm-atlas-7or.pages.dev/models/laguna-xs-2/ https://llm-atlas-7or.pages.dev/models/laguna-xs-2/ Tue, 28 Apr 2026 00:00:00 GMT Poolside · Sparse MoE · 33B total, 3B active (9.1% active) · 128K context. Poolside's open-weight agentic coding MoE uses mixed sliding-window/global attention and activates only 3B of 33B parameters per token. https://llm-atlas-7or.pages.dev/models/deepseek-v4-flash/ https://llm-atlas-7or.pages.dev/models/deepseek-v4-flash/ Fri, 24 Apr 2026 00:00:00 GMT DeepSeek · Sparse MoE · 284B total, 13B active (4.6% active) · 1M context. DeepSeek's efficient V4 preview keeps the million-token architecture while reducing the MoE scale to 284B parameters and 13B active parameters. https://llm-atlas-7or.pages.dev/models/deepseek-v4-pro/ https://llm-atlas-7or.pages.dev/models/deepseek-v4-pro/ Fri, 24 Apr 2026 00:00:00 GMT DeepSeek · Sparse MoE · 1.6T total, 49B active (3.1% active) · 1M context. DeepSeek's flagship V4 preview scales to 1.6T parameters and introduces compressed sparse attention plus manifold-constrained hyper-connections for million-token contexts. https://llm-atlas-7or.pages.dev/models/ling-2-6-1t/ https://llm-atlas-7or.pages.dev/models/ling-2-6-1t/ Thu, 23 Apr 2026 00:00:00 GMT InclusionAI · Sparse hybrid · 1T total, 63B active (6.3% active) · 256K context. Ling 2.5 successor that keeps the Lightning Attention and MLA hybrid stack while adding an MTP layer for multi-token prediction. https://llm-atlas-7or.pages.dev/models/tencent-hy3-preview-295b-a21b/ https://llm-atlas-7or.pages.dev/models/tencent-hy3-preview-295b-a21b/ Thu, 23 Apr 2026 00:00:00 GMT Tencent · Sparse MoE · 295B total, 21B active (7.1% active) · 256K context. Tencent Hunyuan's 295B sparse MoE preview combines a dense first FFN layer, GQA with QK-Norm, and one extra MTP layer for faster generation. https://llm-atlas-7or.pages.dev/models/qwen3-6-27b/ https://llm-atlas-7or.pages.dev/models/qwen3-6-27b/ Wed, 22 Apr 2026 00:00:00 GMT Alibaba Cloud · Dense hybrid · 27B parameters · 256K context. Dense Qwen3.6 model that keeps the Qwen3.5-style Gated DeltaNet/Gated Attention hybrid stack while replacing MoE blocks with dense FFNs. https://llm-atlas-7or.pages.dev/models/xiaomi-mimo-v2-5-310b/ https://llm-atlas-7or.pages.dev/models/xiaomi-mimo-v2-5-310b/ Wed, 22 Apr 2026 00:00:00 GMT Xiaomi · Sparse omnimodal MoE · 310B total, 15B active (4.8% active) · 1M context. Omnimodal sparse MoE model that extends the MiMo-V2-Flash backbone with vision and audio encoders. https://llm-atlas-7or.pages.dev/models/xiaomi-mimo-v2-5-pro-1-02t/ https://llm-atlas-7or.pages.dev/models/xiaomi-mimo-v2-5-pro-1-02t/ Wed, 22 Apr 2026 00:00:00 GMT Xiaomi · Sparse MoE · 1.02T total, 42B active (4.1% active) · 1M context. Trillion-parameter sparse MoE model for long-horizon agentic and software-engineering tasks. https://llm-atlas-7or.pages.dev/models/kimi-k2-6/ https://llm-atlas-7or.pages.dev/models/kimi-k2-6/ Mon, 20 Apr 2026 00:00:00 GMT Moonshot AI · Sparse MoE · 1T total, 32B active (3.2% active) · 256K context. Native-multimodal K2.5 successor that keeps the same 1T sparse MoE backbone while targeting stronger long-horizon coding, design, and agent orchestration. https://llm-atlas-7or.pages.dev/models/qwen3-6-35b-a3b/ https://llm-atlas-7or.pages.dev/models/qwen3-6-35b-a3b/ Wed, 15 Apr 2026 00:00:00 GMT Alibaba Cloud · Sparse hybrid · 35B total, 3B active (8.6% active) · 256K context. Compact Qwen3.6 open-weight MoE that keeps the Qwen3.5 hybrid Gated DeltaNet/Gated Attention recipe while activating only about 3B parameters. https://llm-atlas-7or.pages.dev/models/glm-5-1/ https://llm-atlas-7or.pages.dev/models/glm-5-1/ Tue, 07 Apr 2026 00:00:00 GMT Z.ai · Sparse MoE · 744B total, 40B active (5.4% active) · 198K context. Post-trained GLM refresh that keeps the GLM-5 backbone intact but targets stronger long-horizon agentic coding. https://llm-atlas-7or.pages.dev/models/gemma-4-31b/ https://llm-atlas-7or.pages.dev/models/gemma-4-31b/ Thu, 02 Apr 2026 00:00:00 GMT Google · Dense · 30.7B parameters · 256K context. Dense Gemma 4 scales the family to a 256K-context multimodal checkpoint without changing the core local-global recipe much.