Expired50% confidenceBenchmarkExact time
The Bonsai ternary version is approximately 6GB at 1.71 bits and retains approximately 97% of the base model's quality; the 1-bit version is 3.9GB and retains approximately 90% of quality
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/20/2026
First Seen
Valid until: 10/18/2026(expired)
Sources
Related Claims
UnverifiedBonsai is equipped with a 4-bit vision tower (approximately 0.5GB), supports a context window of approximately 262,000 tokens, and retains full multimodal agent capabilities72% similarUnverifiedBonsai compresses a 27B model from 54GB at full precision down to 3.9GB71% similarUnverifiedTernary Bonsai is a family of 1.58-bit language models where each weight has only three values (1, 0, or -1), making them approximately 9 times smaller than standard 16-bit models; the 8B model requires only 1.7 GB of memory and can exceed 100 tokens per second.70% similarUnverified三值化权重每个权重仅携带约1.58 bits信息量（log₂3≈1.585），相比FP32的32 bits信息密度下降约20倍69% similarUnverifiedTencent Hunyuan released 1-bit and 4-bit quantized versions of Hunyuan3 (Hi3); the original model has a total of 295B parameters, BF16 weights approaching 598G, and the 1-bit version, after compression, can run on a single GPU with 96G of VRAM.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/569980API
curl https://kongchang.com/api/v1/knowledge/claims/569980MCP
get_claim(id=569980)