88Z AI, 29 April — iFlytek has officially launched the Spark X2-Flash large language model and opened its API. The model uses a MoE architecture with 30 billion total parameters, supports a 256K long context window, and was trained on Huawei Ascend 910B domestic computing clusters. It shows significant improvements in agent and coding capabilities, with performance close to trillion-parameter models, token costs less than one-third of mainstream models, and integration with platforms like AstronClaw and Loomy, com
At a glance
- Spark X2-Flash uses MoE architecture, trained on Huawei Ascend 910B, supporting 256K context
- Performance approaches trillion-parameter models with token costs under one-third of mainstream ones
The update
iFlytek's open platform has officially launched the Spark X2-Flash large language model, simultaneously opening its API service. This model adopts a Mixture of Experts (MoE) architecture with a total of 30 billion parameters and supports an ultra-long context window of up to 256K. Its training was completed entirely using Huawei Ascend 910B domestic computing clusters. In terms of performance, Spark X2-Flash achieves significant leaps in agent interaction and code generation capabilities, with tested results approaching the level of trillion-parameter large models. Meanwhile, its token consumption cost is less than one-third of that of mainstream large models. Currently, the model has been integrated into platforms such as AstronClaw and Loomy, and is compatible with major Agent development frameworks like OpenClaw, providing developers with efficient and low-cost AI solutions.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 讯飞开放平台Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.