Start now
KIMISEEK / NEWS

Model news

Updates on Chinese AI models and their practical impact on API integration, model selection, and cost.

  1. open_model

    Reflection AI Releases Open-Weights Beam 501B Model: Apache 2.0 Weights and Two AWS Deployment Paths

    According to an AWS blog post published on October 16, 2026, Reflection AI has introduced Beam, an open-weights model aimed at agentic coding and long-horizon tasks. It has 501B total parameters and 23B activated parameters, a reported effective context of 1 million tokens, weights released under Apache 2.0 via Hugging Face in BF16, FP8 and NVFP4 checkpoints, and two deployment paths: SageMaker JumpStart and HyperPod.

  2. flux

    FLUX.3 Image Klein Open Weights Out With Day-One Diffusers Support: Layout-Box Control and Developer Trade-offs

    According to a Hugging Face blog post published on October 15, 2026, Black Forest Labs' FLUX.3 Image Klein shipped as two open checkpoints and was available in Diffusers on day one. The model supports text-to-image, multi-reference image-to-image and bounding-box layout control under the same weights; the distilled checkpoint generates a 1024x1024 image in about 2 seconds on an H100, with NVFP4 and FP8 single-file weights and LoRA training scripts also provided.

  3. open_model

    Fish Audio releases Drama 3: 35B MoE open-weight TTS, starting at 48GB VRAM

    According to a MarkTechPost report dated October 14, 2026, Fish Audio released Drama 3, a mixture-of-experts text-to-speech model with 35B total parameters and about 3B activated per token, with weights open on Hugging Face under a community license capping commercial use at $500,000 in revenue, and FP8 self-hosting requiring more than 48GB of NVIDIA VRAM.

  5. domestic_model

    USTC-Affiliated Baize Tongjing Releases Abductive World Model AWM and Edge Engine BAIZ-RUN

    According to an October 2026 report on the WeChat public account "Machine Heart", USTC-affiliated Hefei-based Baize Tongjing proposed the Abductive World Model AWM, which it says can learn entities, dynamics and relations from unlabeled video, with a paper and code released; the companion edge engine BAIZ-RUN reported quantitative gains only on AMD edge integrated graphics and the SSV2 dataset. Neither version appears in the current KimiSeek catalog.

  6. open_model

    Nace.AI Releases Drex 1.5 as Open Weights: 9B Decision Model Scores Options in a Single Forward Pass, Hits 93.4% on 128K Long Documents

    According to a MarkTechPost report dated October 9, 2026, Nace.AI has released Drex 1.5, an 8.95B dense decision model, as open weights. It reads a state and a typed question, then returns probabilities for each option without generating text. The report says it scores 58.08 on Decision Index 0.3.1 and can run locally in bf16 on a single GPU or via a roughly 9.5 GB quantized version.

  7. domestic_model

    Westlake Robotics Upgrades General Brain WR1: Two-Robot Folding and Three-Room Chores Demo

    According to an October 10, 2026 report, Hangzhou-based Westlake Robotics unveiled an upgraded version of its general-purpose brain model WR1, combining a world model with a VLA architecture. In a continuous single-take demonstration, the system carried out a sequence of tasks: picking up laundry, two humanoid robots collaboratively folding trousers, retrieving corn and placing it in an air fryer, tidying up plush toys, and making a bed, spanning the kitchen, living room, and bedroom. The report claims the dexterous-hand autonomous clothes folding is the world's first such demonstration.

  8. stepfun

    Step 5 Preview Tops OpenRouter Trending: 600B-Parameter Sparse MoE, Official Open-Source Date Set for October 15

    According to a Leiphone report dated October 10, 2026, StepFun's Step 5 Preview topped OpenRouter's Trending list and ranked second globally on the daily list, using a sparse MoE architecture with 600 billion total parameters and about 27 billion activated per token; the company said it would open-source the model on October 15, and its Artificial Analysis composite intelligence index is 44.

  9. qwen

    Qwen releases Qwen-Image-2.1-Turbo: a 7B image model with 8-step generation, research license and hosted API side by side

    According to a MarkTechPost report on October 9, 2026, Alibaba's Qwen released Qwen-Image-2.1-Turbo, which compresses the default 40-step denoising of Qwen-Image-2.1 to 8 steps, retains the 7B generator and Qwen3-VL 8B text encoder, and offers a hosted API at 0.1 yuan per image. The weights are under a research license, and commercial self-hosting requires separate authorization.

  11. open_model

    JetBrains Open Sources Mellum2.1: 12B MoE Coding Agent Model With Apache 2.0 Weights and Quantizations

    According to an October 8, 2026 report by MarkTechPost, JetBrains released Mellum2.1 under the Apache 2.0 license, namely Mellum2.1-12B-A2.5B-Thinking: a 12B total-parameter, 2.5B-active-parameters-per-token MoE thinking model with the same architecture as the previous generation, where the improvements come mainly from reinforcement learning in real code repositories, with the report stating SWE-bench Verified rose from 2.0 to 47.0.

  12. vidu

    Shengshu Technology releases Vidu Q4 Preview: up to 15 reference images, 2K/4K output, first-release pricing from 0.09 yuan/second

    According to Machine Heart (Jiqizhixin), Shengshu Technology has introduced Vidu Q4 Preview, focused on character performance, camera language and complex visual effects, supporting up to 15 reference images, 3 reference audio clips and 2K/4K output, with first-release pricing from 0.09 yuan per second; the report says image-to-video speed is about 2x that of comparable models. The performance conclusions come from that outlet's own testing. This version is not in the live KimiSeek catalog.

Model news | KimiSeek