Unverified50% confidenceFactExact time
Xingchen 4.0 is built on a data foundation of hundreds of trillions of tokens, adopts architectural designs including MLA, MTP, and MHC, and undergoes staged long-context training from 4K to 32K, 128K, and 256K.
1
Sources
50%
Confidence
Long-term
Relevance
9/24/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe context window of LLMs has expanded from the early 4K tokens to hundreds of thousands of tokens, enabling models to maintain state across complex multi-step tasks66% similarUnverifiedCombined with interpolation techniques such as YaRN and Dynamic NTK, models trained with a context length of 4K can be extended to context windows of 128K or even 1M.65% similarUnverifiedThis configuration achieves three optimizations—262K ultra-long context, KV Cache Q4 quantization, and MTP speculative sampling—pushing 24GB VRAM utilization to 99%63% similarUnverified位置编码外推方案可将模型有效上下文从最初的4K、8K扩展至128K乃至百万级token63% similarVerifiedThe context window of large language models has expanded from 4K to the million-token level.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/946992API
curl https://kongchang.com/api/v1/knowledge/claims/946992MCP
get_claim(id=946992)