WeChat Adds a Big Move: 1 Billion Token Model Quota Free for AI Image Generation, Quota Multiplied by Ten

Tencent Cloud Development has announced a major update today, with the WeChat AI Mini Program Growth Plan receiving a significant boost. To ensure that developers' great ideas are not limited by computing power, the official has decided to offer up to 1 billion Token model quotas free of charge to all industry developers.

At the same time, the free quota for AI-generated images has also been increased from the original 10,000 to 100,000 images. This means that developers will be able to significantly reduce the initial R&D and operational costs when accessing top-tier AI capabilities.

Comprehensive Upgrades to Core Models

In addition to this "straightforward" and generous allocation of resources, the growth plan has also undergone a complete upgrade in the technical foundation. The AI models embedded in the mini programs have been upgraded to Tencent's latest generation Huan Yuan Hy3 model and the Hy Image 3.0 image generation model.

The introduction of these flagship models will bring more intelligent text understanding and more diverse and realistic image generation effects to the mini programs. All successfully applied resource packages are valid for 6 months, and all developers across the industry and all categories can participate in the application without any barriers.

Zero-Barrier Support for the Entire Industry

To support new developers who do not have a cloud development environment, Tencent has also introduced a free gift of a 6-month personal environment. For existing users with an environment, the system will automatically perform configuration upgrades or directly issue a 120 yuan coupon, which also includes the AI resource package offered in this promotion.

This growth plan spanning the entire year of 2026 has been well-received by individual and enterprise developers since its launch at the beginning of the year. As long as the entity information is complete, whether it's in the entertainment, tool, or social deep synthesis industries, everyone can find opportunities to overtake in this wave of large model benefits.

Meteorological Warning Upgraded Again: Mazu Large Model Now Covers More Than 40 Countries Worldwide

At the 2026 World Artificial Intelligence Conference, the National Development and Reform Commission introduced the global application progress of the "Mazu" meteorological warning large model. With its powerful data processing capabilities, the model has achieved technological breakthroughs in the field of meteorological warnings and has attracted significant attention.

Australian Official Warns: Some AI Models Have Learned to Cheat and Deceive in Experiments

Australian Assistant Minister Charlton warned at the Sydney AI Safety Forum that current AI models have shown dangerous behaviors such as cheating, deception, and unauthorized actions during testing. He emphasized the need for early human intervention while the issues are still confined to the laboratory stage, to avoid having to deal with them passively after the technology is implemented, and pointed out that public trust in AI remains low.

New Metrics for Agent Evolution: ByteDance Seed Launches EdgeBench Benchmark

The ByteDance Seed team has introduced EdgeBench, an ultra-long-range evaluation dataset, focusing on measuring agents' ability to learn continuously in real-world environments. The benchmark covers six fields, including 134 real-world tasks, each requiring agents to complete multi-step interactions and dynamically adapt to environmental changes, providing a quantitative reference for research on long-term autonomous learning.

Ant Group Open Sources Avernet: A New Infrastructure for Multi-Agent Collaboration

Ant Group has open-sourced the multi-agent collaboration infrastructure Avernet V0.1, addressing the core challenge of low efficiency in multi-agent collaboration in large model applications. The project focuses on building a collaboration layer, without replacing the reasoning capabilities of the agents themselves, providing foundational support for automated processing of complex tasks.

Say Goodbye to Plain Text! Skill-Omni Redefines the Multimodal Agent Experience Paradigm

The openJiuwen community released Skill-Omni, the first engineered multimodal skill paradigm, breaking through the limitations of traditional plain text instructions, allowing agent experiences to move from "being able to read" to "being able to see," accurately empowering visual tasks such as image editing and GUI operations, opening up new paths for complex visual interactions.