DeepSeek redirects all V4-Pro API traffic to V4.1-Flash with Flash rates

Cambricon announced the successful completion of adapting the DeepSeek-V4.1-Flash model as part of the Day-0 program based on the open inference stack vLLM. This means that the recently released model can run stably on Cambricon accelerators on the same day DeepSeek published the checkpoint.

This announcement highlights the synergy between chip and software: simultaneous availability of the model and the chip, as well as the maturity of Cambricon's approach combining NeuWare and vLLM. This is not a repetition of information about the general availability of the model itself.

DeepSeek-V4.1-Flash represents the smallest variant of the new DeepSeek architectural line. It is a mixture-of-experts model with 552 billion parameters, which activates about 8 billion parameters on the input side and 16 billion on the output side, and also possesses built-in multimodal vision understanding. DeepSeek notes the use of the Causal-Encoder-Decoder structure and aggressive KV cache compression, which reduces HBM requirements by approximately one quarter and SSD requirements by one eighth compared to the previous generation, while the cache size is hundreds of times smaller than the first version.

These memory savings are critically important for accelerator manufacturers who aim to control inference latency, DRAM volume, and storage levels on dense servers that are already straining HBM budgets.

For its part, Cambricon engineers used the Torch-MLU-Ops unified operator library to accelerate structures such as Engram and Compressor, and wrote optimized kernels in BangC for sparse and compressed attention hotspots. Within vLLM, Cambricon reports support for five-dimensional mixed parallelism, covering tensor, pipeline, sequential, batch, and expert parallelism, with communication and computation overlap, low-precision quantization, and prefix and decoding splitting to boost end-to-end throughput. This stack operates on NeuWare, the company's long-term software platform.

Several days ago, Cambricon joined the PyTorch Foundation as a platinum member, taking a seat on the governing board alongside other major hardware and cloud service developers, signaling deeper investments in the open learning and inference ecosystem.

Cambricon now supports Day-0 inference for five major Chinese model lines: GLM, DeepSeek, Qwen, Kimi, and MiniMax, representing Zhipu AI, DeepSeek, Alibaba, Moonshot AI, and MiniMax, respectively. Caixin Global specifically noted the ability to include Cambricon on the same day as an indicator of shortening internal development cycles for models and chips in China—a shift from multi-month closed ports to readiness for launch on public frameworks that are already standardized by operators.

The practical takeaway for cluster operators is co-availability: when a cutting-edge open model is released, Cambricon clients can use the documented vLLM recipe instead of waiting for a closed fork. The next test for the software ecosystem will be whether this Day-0 scheme applies to larger DeepSeek variants and multimodal serving mixtures.

DeepSeek, a Chinese company, has introduced V4.1-Flash, its latest artificial intelligence model. This new technology was presented as the smallest within its family of models and promises to offer more agile responses, greater processing capacity, and lower operational costs.

This launch comes as DeepSeek prepares for an Initial Public Offering (IPO) on the STAR market in Shanghai, as reported by Reuters. Furthermore, the company is implementing V4.1-Flash, replacing previous versions of its models.

Despite being the smallest model in the new line, V4.1-Flash has 552 billion parameters, which are the components used by the AI to process data and generate outputs. However, during each task, it uses only a fraction of these resources, specifically 8 billion parameters to receive information and 16 billion to formulate responses.

The practical objective for the user is to allow the model to operate with greater speed and handle a higher volume of requests without demanding excessive computational resources. DeepSeek also assures that the system was designed to facilitate the development of even larger models in the future.

The company highlighted that V4.1-Flash features 'native visual understanding,' allowing it to interpret visual content, such as images. Tests conducted by various entities indicated that this new model surpasses V4-Pro in terms of performance, cost, speed, and total execution time.

A significant change lies in how the model stores data during an operation, which historically could increase the costs of long-duration AI services. Compared to the previous generation, DeepSeek stated that V4.1-Flash requires fewer resources, which, according to the company, can drastically reduce expenses, particularly in AI agents and systems that perform tasks with little human intervention.

V4.1-Flash is already accessible through the DeepSeek API, which allows the integration of the model into other applications. The V4-Flash and V4-Flash-Vision-Exp models have been discontinued, and their calls are temporarily being redirected to the new system.

More Information:

V4-Pro will undergo a gradual withdrawal. Starting September 14, 2026, all requests directed to it will be automatically sent to V4.1-Flash, applying the charge corresponding to the new model. This process will continue until the launch of V4.1-Pro.

The new API rates will take effect on September 10. DeepSeek establishes different tariffs for peak and off-peak periods; outside of peak hours, the price will be half the maximum rate. The company emphasized: 'V4.1-Flash allows serving more users at a lower cost. We are passing this savings on to you.'

Additionally, DeepSeek plans to collaborate with the open-source community to expand support for the model and investigate other implementation methodologies. The report on the launch of the new model was initially published in Olhar Digital.