New Quantum Encryption Method Prevents Copying Encrypted Messages

Following statements about the aspiration to make Pangu a world-class leader, Yu Chengdong unveiled openPangu-2.0-Pro. Huawei has released the source code for openPangu-2.0-Pro along with weights, inference code, and a technical report. This model is the first in the series of 500 billion parameters to be fully trained on hardware other than NVIDIA, strengthening the domestic computing ecosystem.

On July 31, Huawei made openPangu-2.0-Pro publicly available. The model features 505 billion total parameters, approximately 180 billion sparse activations, and a context window of 512K. It was trained on Ascend NPU based on 34 trillion tokens. A key distinction is the fully functional open-source nature: unlike many other open-source models that only provide weights, openPangu-2.0-Pro offers a completely reproducible chain from training on Ascend to native Ascend inference. This is the first frontier-level model with over 500 billion parameters fully trained on non-NVIDIA hardware. Seven components were released, including pre- and post-training code. Yu Chengdong explained the choice of 505 billion parameters by stating that computational power is largely utilized by other domestic enterprises, leaving limited resources for Huawei.

The model's architecture is optimized for resource constraints and incorporates hybrid DSA plus SWA attention, four-thread mHC residuals, three-head MTP self-speculation, and the Muon optimizer. With a 128K context window, the minimum TPOT latency is 9.55 milliseconds, and single-card throughput reaches 1326 tokens per second, enabling real-time operation at the edge. The 512K context window with sparse activation allows the model to comprehensively analyze data and perform computations efficiently: it can read an entire set of contracts or a factory's annual log in a single pass while maintaining cost control. The model is oriented towards long-context agent tasks, becoming a digital employee capable of multi-step decision-making, planning, and tool invocation, marking a transition from simple chat to full functionality for devices, vehicles, and industrial control. The report frankly admits that openPangu-2.0 still lags behind leading models in complex real-world software engineering tasks, reflecting a long-term strategy.

The Pangu model acts as an industry expert in AI for industry, utilizing the L0 foundation, L1 industry, and L2 scenario architecture. State-level models operate entirely locally. For instance, Pudong Development Bank created the Pu Hui Cloud Cang warehouse bank, which improved cargo detection by 5–10% and reduced development time from months to days. PetroChina Kunlun achieved a 40% increase in sub-millimeter defect elimination, and manufacturers began saving 20 tons of fuel daily per furnace. Intelligent driving benefits from understanding trajectories within the 512K window. Guangyao reduced early drug discovery costs by 70%, and Hanyu improved production decision-making efficiency by 90%. Tianjin Energy achieved 100% heating balance while reducing energy consumption by 10%. Small businesses benefit the most: now district hospitals or local agricultural institutes can own industry agents.

The release was deliberate: the unveiling of seven components began on June 30 at HDC 2026. Initially, a 92-billion parameter Flash model was released, followed a month later by the heavier Pro version, allowing for the provision of training methodologies and engineering knowledge, rather than just releasing a repository. This approach echoes the view of Jian Zhengfei on Apple: open collaboration focusing only on the most advantageous part. In the age of AI, ecosystem victory precedes the performance of a single machine. Open channels guide developers into the Ascend plus Pangu river, and as users grow, this river expands. openPangu-2.0-Pro serves as a 'keystone' for the domestic community: when leading players collectively disseminate their capabilities, lower-level applications flourish, and Chinese enterprises can form a unified force on an independent stack for the first time. For a company under intense pressure, handing over the heaviest core technology to everyone is the highest form of trust.

The DeepSeek V4 Flash model demonstrated massive demand, consuming 8 trillion tokens in a single day on the OpenCode tool. This volume exceeded the average daily usage of the OpenRouter platform, which is around 6.6 trillion tokens. This surge in usage reflects a fundamental shift in AI interaction methods: where tasks used to be simple question-and-answer exchanges, agents now perform more complex operations such as reading files, modifying code, running programs, and retrying after errors, which significantly increases token consumption.

An example of such complex interaction was shown by a user who created a single-page 3D game project using Claude Opus 5. This process resulted in one HTML file that required 690 million tokens and cost nearly 3000 yuan.

With the advent of agent-driven tasks, the evaluation criterion for models has shifted from the quality of a single response to the cost of completing a long-running task. For instance, one million output tokens cost 2 yuan with DeepSeek, whereas the same operation with Claude Opus 4.8 cost $25 USD, equivalent to approximately 170 yuan, demonstrating an 85-fold price difference. Although the V4 Flash Max model scored 50 points compared to 56 for Opus Max in the Artificial Analysis Intelligence Index v4.1, the cost to pass this test was $72.02 for V4 Flash versus $3752.55 for Opus, with Opus gaining 6 extra points at roughly a 52-fold increase in call costs.

V4 Flash cannot yet prove absolute superiority, but if both models appear in the same candidate list, the 85-fold price difference could change the standard selection. When a model only slightly outperforms its competitor but costs dozens of times more, the mathematics of choice no longer favors the more expensive option.

The first to feel the impact of this trend are not the weakest models, as smaller models remain suitable for routing, and flagship models continue to handle the most complex tasks. Mid-tier models are the most vulnerable—they are stronger than V4 Flash but cost dozens of times more and are not powerful enough to solve the tasks that V4 Flash handles. This is DeepSeek's 'killer line': models do not necessarily need to be the strongest; they just need to be good enough to perform most tasks, leveraging a price gap of almost two orders of magnitude to displace more expensive options from the default list.

In terms of pricing, Opus costs $25 per million output tokens, Fable 5 costs $50, and Chinese models include Qwen 3.8 Max priced at 36 yuan, Kimi K3 at 100 yuan, and GLM 5.2 at 28 yuan.

The ability to sell Mutai at mineral water prices is due to the direct inference cost per token, which depends on activated parameters, accuracy, attention, KV cache, output length, hardware utilization, and parallelism. V4 Flash has 284 billion total parameters but activates about 13 billion per token by using hybrid attention, low-precision weights, and cache optimization. With a context of one million tokens, the inference computational load is about 10% of V3.2, and the KV cache size is about 7%. The V4 series overhauled the attention mechanism, implementing CSA compress-then-focus and HCA full-browse mechanisms, and utilizes DeepSeekMoE routing, directing each token to a small subgroup of experts. Commercial pricing and service costs follow these technical features. Thus, the 'killer line' redefines the boundary of development: the combination of performance and price becomes the decisive factor, where the cost of task execution is the determining metric.