Unverified50% confidenceReleaseExact time
Recursive released Repo65B preselected weights, a 65-billion-parameter dense model offered in base and instruction versions plus an experimental low-precision version, under the Apache 2.0 license, with official support for up to 1M context.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/8/2026
First Seen
Valid until: 2/6/2027
Sources
Related Entities
Related Claims
UnverifiedDecision Models use the laya model under the hood, which is 449 MB in size and licensed under Apache-2.0.68% similarUnverifiedRecursive released a local coding MoE model called Flint, with 40 billion total parameters, only 2.5 billion parameters activated per inference, context length support up to 512K, and an official recommendation of 256K for daily use.67% similarUnverifiedThe model has 550 billion parameters, activates 55 billion parameters, uses a 550B-A55B architecture, and is an open-source model.66% similarUnverifiedA 125-billion-parameter model can run on 12G of VRAM, relying mainly on model quantization and a mixture-of-experts (MoE) architecture, achieved by activating only part of the expert parameters, quantizing weights to 4bit or lower, and dynamically loading inactive parameters from memory or SSD on demand.61% similarUnverifiedFor a 7-billion-parameter model as an example, LoRA may only need to update 0.1% to 1% of the parameters, reducing the required VRAM from hundreds of GB to a level runnable on consumer-grade GPUs such as the RTX 4090.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1147352API
curl https://kongchang.com/api/v1/knowledge/claims/1147352MCP
get_claim(id=1147352)