Ant Group Launches Ling 3.0 Flash with 5.1 Billion Tokens Activated per Token

Ant Group's AI organization, inclusionAI, has announced the Ling 3.0 Flash, which activates only 5.1 billion tokens out of a total of 124 billion parameters. The capacity of the FP8 quantized version is reported to be approximately 128GB, half that of the BF16 version. The BF16 original version and FP8 quantized version of Ling 3.0 Flash have been uploaded to Hugging Face and ModelScope under the MIT license and can be distributed via SGLang or vLLM. This model adopts a Mixture-of-Experts (MoE) structure, calling only a subset of expert networks based on the input to reduce computational load. It supports a base context window of 256K and can scale up to 1M tokens. It aims for a peak inference speed of 1000 tokens per second in high-frequency tasks and a first token response time of under 100ms. According to the official documentation, Ling 3.0 Flash applies a hybrid linear attention structure with KDA and MLA layers cross-deployed at a 5:1 ratio. Ling 3.0 Flash will be available for free from 1 AM on July 23 to 12:59:59 AM on August 7, Korean time, and in August, it will be offered at a 75% discount, with the regular price set at 0.40 yuan per million input tokens and 1.20 yuan per million output tokens.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.