Unverified50% confidenceReleaseExact time
The new lab ReflectionAI released the open-source model BIM, with about 501 billion total parameters and 23 billion active parameters, adopting a MoE architecture and fully pretrained from scratch.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/22/2026
First Seen
Valid until: 11/5/2026
Sources
GPT-Next Leaked, Gemini 4 Argon Speeds Up, Fable 5.5 May Launch Today
bilibiliAI-宇宙10/8/2026
Related Entities
Related Claims
VerifiedThe concept of Foundation Model was formally proposed by Stanford HAI in 2021, referring to general-purpose models pre-trained on ultra-large-scale data that can be transferred to various downstream tasks with a small amount of labeled data.63% similarUnverifiedBIM (Building Information Modeling) is a working methodology that uses three-dimensional digital models as a carrier to integrate information across the entire building lifecycle.62% similarUnverifiedThe paper conducts pretraining evaluation on the Llama-3 family 8B model, with a training scale of 160 billion tokens, using bfloat16 for the output projection layer and employing compiled cross-entropy.58% similarUnverifiedByteDance's Seed team plans to train a foundation large model with more than 50 trillion parameters; the project is currently in the early validation stage and insists on native pre-training while rejecting the distillation approach.58% similarUnverified在 LLM 输出需要被结构化解析的场景中，引入 valibot 这类轻量校验库可以在运行时捕获模型返回格式异常。57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1087818API
curl https://kongchang.com/api/v1/knowledge/claims/1087818MCP
get_claim(id=1087818)