MoE

China Telecom AI's Xing4.0-29B-A4B (formerly the TeleChat line) is a mixture-of-experts model with 29B total and 4B active parameters and a native 256K context window, extensible to 512K. The company claims it is the first model of this scale trained entirely on Ascend NPUs with MindSpore. Community GGUF quantizations from Venastine-Research are already available.

inclusionAI has released Ling 3.1 Flash, a hybrid reasoning mixture-of-experts model with 560B total and 25B active parameters and a 262K-token context window. It is listed as free on OpenRouter through NovitaAI. No benchmark scores have been published on the listing.

Ai2 released Olmo-core 3, an open training framework for large mixture-of-experts models. The company reports about 2.7× the throughput of its earlier FSDP-based implementation and benchmarks up to 1.2 trillion total parameters on 512 NVIDIA B300 GPUs. It is the infrastructure for Ai2's next MoE-based Olmo, not a model release.

Xiaomi released MiMo-V2.6-Pro, a 1.02T-parameter mixture-of-experts model with 42B active parameters, which debuted as the top-scoring open-weights model on Artificial Analysis' Intelligence Index (46). The company claims the model's RL training run cost roughly $2.6M and completed in 130 hours.

Xiaomi's MiMo team released MiMo-V2.6-Flash-RL, an efficiency-tier checkpoint in the MiMo-V2.6 series featuring a 309B-parameter (15B active) Mixture-of-Experts architecture, 1M-token context, and native support for text, image, video, and audio. The model uses a single mixed reinforcement learning run across coding, agentic, visual, and cybersecurity tasks rather than domain-specific training.

Yandex has released AliceAI-Foundation-80B-A3B-Base, an 80-billion-parameter hybrid MoE base model with 3 billion active parameters per token and a 262,144-token context window. The model was trained fully from scratch and, according to Yandex, outperforms larger open-source models on Russian-language factual and educational benchmarks.

China Telecom Artificial Intelligence Technology has released Xing4.0-29B-A4B, a 29-billion-parameter mixture-of-experts model with only 4B parameters active per token and native 256K context. The company claims it is the first model of this scale trained entirely on Huawei's Ascend NPU platform using the MindSpore framework.

InclusionAI has released Ling 3.0 Flash VL, a vision-language extension of its 124B total-parameter, 5.5B active Mixture-of-Experts model. The model adds native image and video understanding, supports a 131K token context window, and is priced at $0.06 per 1M input tokens and $0.18 per 1M output tokens via OpenRouter.

Alibaba's Qwen team released Qwen3.8-2.4T-A95B on August 12, 2026, the open-weight version of Qwen3.8-Max and the first Qwen-Max-class model made publicly available. The 2.4 trillion-parameter mixture-of-experts model activates only 95 billion parameters per token and supports context windows up to 1 million tokens.

Alibaba's Qwen team has released Qwen3.8-Flash-Next, an open-weight model with 125B total parameters (6B activated) that previews architectural changes planned for Qwen4, including a new sparse attention mechanism and n-gram embeddings. The model natively supports 262,144 tokens of context, extensible to 1 million.

Alibaba's Qwen team has released Qwen3.8-2.4T-A95B-FP8, an open-weight, FP8-quantized MoE model with 2.4 trillion total parameters and 95 billion activated per token. It natively supports 262,144 tokens of context, extensible to 1,010,000, and forms the base for the hosted Qwen3.8-Max API.

NVIDIA released the full-precision BF16 reference weights for Nemotron 3.5 Lightning, a 30B-parameter Mixture-of-Experts model with only 3B active parameters and support for up to 1 million tokens of context. The model uses a hybrid Mamba-2, MoE, and Attention architecture and is licensed under OpenMDW-1.1 for commercial use.

Alibaba's Qwen team has released Qwen3.8-2.4T-A95B, a 2.4-trillion-parameter mixture-of-experts model activating 95B parameters per token. The company claims it is the first open-weight release to reach Qwen-Max-class performance, with a hosted variant, Qwen3.8-Max, offering 1M-token context and vision input.

NVIDIA has published NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4, a 30-billion-parameter Mixture-of-Experts model with only 3B active parameters, a hybrid Mamba-2/MoE/Attention architecture, and support for up to 1 million tokens of context. The NVFP4-quantized checkpoint is designed to run on a single DGX Spark (GB10) or H100 GPU.

A dense wave of open-weight model releases—including Thinking Machines' first model Inkling, Poolside's Laguna S2.1, and Moonshot AI's Kimi K3—signals that consolidation predictions for AI labs have not materialized. Licensing terms, particularly Kimi K3's noncommercial agreement requirement, are emerging as a new front in US-China AI policy.

Thinking Machines Lab has released Inkling Small, an open-weight multimodal mixture-of-experts model with 12B active parameters out of 276B total and a 524K token context window. The model targets reasoning, coding, agentic workflows, and multilingual use cases at $0.58 per 1M input tokens and $1.44 per 1M output tokens.

Moonshot AI has released open weights, a technical report, and supporting infrastructure for Kimi K3, a model that claims 2.5x more intelligence per unit of compute. Independent testing found notable gaps in cybersecurity and math performance compared to Western frontier models.

NVIDIA has released Nemotron-Labs-3-Puzzle-75B-A9B, a compressed variant of Nemotron-3-Super that reduces the model from 120.7B total/12.8B active parameters to 75.3B total/9.3B active parameters. According to NVIDIA, the model achieves approximately 2× higher server throughput on a single 8×B200 node and increases sustainable 1M-token single-H100 concurrency from 1 request to 8 requests while maintaining strong accuracy across benchmarks.

NVIDIA released Audex-30B-A3B, a unified audio-text model built on the Nemotron-Cascade-2-30B-A3B backbone. The model handles audio understanding, speech recognition and translation, text-to-speech, audio generation, and speech-to-speech while supporting up to 1M token context length.

Unsloth has released optimized GGUF quantizations of DeepSeek-V4-Flash, a 284B parameter Mixture-of-Experts model that activates 13B parameters and supports 1 million token context windows. The Q8 quantization (UD-Q8_K_XL) runs at 162GB with claimed lossless precision, only 7GB larger than the Q4 variant.

Tencent has released Hy3, a 295-billion parameter Mixture-of-Experts model with 21 billion active parameters and 3.8 billion MTP layer parameters. The model features a 256K context window and is released under Apache 2.0 license, with pricing not yet disclosed.

NVIDIA released Nemotron-Labs-TwoTower-30B-A3B-Base-BF16, a block-wise diffusion language model that generates text by denoising blocks of tokens in parallel rather than sequentially. According to NVIDIA, the model achieves 2.42× the wall-clock generation throughput of its autoregressive baseline while retaining 98.7% of aggregate benchmark quality.

Alibaba's Qwen team released Qwen-AgentWorld-35B-A3B, a 35 billion parameter language world model designed for agentic environment simulation. The model covers seven domains—MCP tool calling, Search, Terminal, Software Engineering, Android, Web, and OS—in a single model with a 262,144 token context window.

NVIDIA released a quantized version of Google DeepMind's DiffusionGemma 26B A4B IT, a multimodal model with 25.2B total parameters (3.8B active) that processes text, image, and video inputs. The NVFP4-quantized model achieves generation speeds exceeding 1,100 tokens per second on NVIDIA H100 GPUs while supporting a 256K token context window.

Moonshot AI has released Kimi K2.7 Code, a 1 trillion parameter Mixture-of-Experts model designed for long-horizon coding tasks. The model features a 256K context window and reduces thinking token usage by approximately 30% compared to its predecessor K2.6.

Cohere Labs has released North Mini Code 1.0, a 30B-parameter sparse Mixture-of-Experts model with 3B active parameters and a 256K context window. The Apache 2.0-licensed model is optimized for agentic software engineering, featuring 128 experts with 8 activated per token, and trained specifically for tool use in coding tasks.

Nvidia has released Nemotron 3 Ultra, a 550-billion parameter mixture-of-experts model with 55 billion active parameters and support for up to 1 million token context windows. The model uses a hybrid Transformer-Mamba architecture and is designed specifically for long-running agentic workflows including agent orchestration, coding agents, and complex enterprise tasks.

NVIDIA Nemotron 3 Ultra is now available on Amazon SageMaker JumpStart with 550 billion total parameters and 55 billion active parameters. The model features a hybrid Transformer-Mamba Mixture-of-Experts architecture and supports context windows up to 1 million tokens, targeting agentic AI workloads.

JetBrains has released Mellum2, a 12-billion parameter Mixture-of-Experts model that activates only 2.5 billion parameters per token. The open-source model is designed for code generation, RAG pipelines, and agent workflows with 2x faster inference than similar-sized models.

StepFun has released Step-3.7-Flash, a 198B-parameter sparse Mixture-of-Experts vision-language model that activates approximately 11B parameters per token. The model supports a 256K context window, native image understanding via a 1.8B-parameter vision encoder, and offers three selectable reasoning levels.

StepFun has released Step 3.7 Flash, a 196B-parameter Mixture-of-Experts model that activates approximately 11B parameters per token. The multimodal model supports a 256K context window and introduces selectable reasoning levels (high/medium/low), priced at $0.20 per 1M input tokens and $1.15 per 1M output tokens.

Tencent released Hy-MT2, a family of multilingual translation models available in 1.8B, 7B, and 30B-A3B (MoE) sizes. All models support translation among 33 languages and follow translation instructions in multiple languages. The 1.8B model can be compressed to 440MB using 1.25-bit AngelSlim quantization.

Zyphra has released ZAYA1-8B, a mixture-of-experts language model with 760M active parameters and 8.4B total parameters. The model scores 89.1% on AIME 2026, competitive with models exceeding 100B parameters, while maintaining efficiency for on-device deployment.

NVIDIA released Nemotron-3-Nano-Omni-30B-A3B, a multimodal large language model with 31 billion parameters that processes video, audio, images, and text with up to 256K token context. The model uses a Mamba2-Transformer hybrid Mixture of Experts architecture and supports chain-of-thought reasoning mode.

NVIDIA released Nemotron 3 Nano Omni, a 31B parameter (30B active, 3B per token) multimodal model supporting video, audio, image, and text inputs. The model features a 256K token context window, reasoning mode with chain-of-thought, and tool calling capabilities.

NVIDIA has released Nemotron 3 Nano Omni 30B-A3B, a multimodal large language model with 31 billion parameters using a Mamba2-Transformer hybrid Mixture of Experts architecture. The model supports video, audio, image, and text inputs with a 256K token context window and includes a dedicated reasoning mode with chain-of-thought capabilities.

NVIDIA has launched Nemotron 3 Nano Omni on Amazon SageMaker JumpStart, a multimodal model with 30 billion total parameters (3 billion active) that processes video, audio, images, and text in a single inference pass. The model features a 131K token context window and uses a Mamba2 Transformer Hybrid MoE architecture combining three specialized encoders.

NVIDIA released Nemotron 3 Nano Omni, a 30B-A3B Mixture-of-Experts model that processes text, images, video, and audio. The model uses a hybrid Mamba-Transformer architecture with 128 experts and achieves 65.8 on OCRBenchV2-En and 72.2 on Video-MME, while delivering up to 9x higher throughput on multimodal tasks compared to alternatives.

Xiaomi has released MiMo-V2.5-Pro, an open-source Mixture-of-Experts model with 1.02 trillion total parameters and 42 billion active parameters. The model supports up to 1 million tokens context length and claims 99.6% on GSM8K and 86.2% on MATH benchmarks.

Alibaba Cloud has released Qwen3.6 Max Preview, a proprietary frontier model built on sparse mixture-of-experts architecture with approximately 1 trillion total parameters. The model supports a 262,144-token context window and features integrated thinking mode for multi-turn reasoning, priced at $1.30 per million input tokens and $7.80 per million output tokens.

DeepSeek has released V4 Pro, a Mixture-of-Experts model with 1.6 trillion total parameters and 49 billion activated parameters. The model supports a 1-million-token context window and costs $1.74 per million input tokens and $3.48 per million output tokens.

DeepSeek has released V4 Flash, a Mixture-of-Experts model with 284B total parameters and 13B activated parameters per request. The model supports a 1,048,576-token context window and is priced at $0.14 per million input tokens and $0.28 per million output tokens.

Tencent has released Hy3-preview, a 295-billion-parameter Mixture-of-Experts model with 21 billion active parameters and a 256K context window. The model scores 76.28% on MATH and 34.86% on LiveCodeBench-v6, with particularly strong performance on coding agent tasks.

Arcee AI has released Trinity Large Preview, a 400B-parameter sparse Mixture-of-Experts model with 13B active parameters per token using 4-of-256 expert routing. The model supports context windows up to 512K tokens and is available with open weights under permissive licensing.

Moonshot AI has released Kimi K2.6, an open-source multimodal model with 1 trillion total parameters (32B activated) and 256K context window. The model achieves 80.2% on SWE-Bench Verified, 58.6% on SWE-Bench Pro, and supports horizontal scaling to 300 sub-agents executing 4,000 coordinated steps.

Alibaba has released Qwen3.6-35B-A3B, the first open-weight model in the Qwen3.6 series. The model features 35B total parameters with 3B activated, a native 262K context window extensible to 1.01M tokens, and achieves 73.4% on SWE-bench Verified using 256 experts with 8 activated per token.