Second model on the Surplus launchpad: $BIT by @LittleBit_llm.
Big models, little bits.
β Today: Qwen3-4B packaged in 4-bit MLX for Apple Silicon, open weights on Hugging Face
β Next: pushing Qwen3 toward 0.55 bits per weight, about 0.28 GB of weights for a 4B model
β After graduation, its own creator fees pay for the GPU that serves it
And 10% of every fee buys back and burns base:0xc52aedec3374422d7510e294cfaa90799595cba3. More launches, more burn.
CA: 0x86af99052db086a66af7d764bb04d3abb81b485f
First LittleBit release: Qwen3-4B in 4-bit MLX for Apple Silicon.
2.1 GB on disk (BF16 is ~8 GB)
~74 tokens/s on an M1 Max
2.4 GB peak memory
Standard 4-bit for now. The sub-1-bit build is still training.
pip install mlx-lm
https://t.co/BDGFMJe3hY
Holders, here is something to do with $IROGENT besides watching it
Open https://t.co/9eUuZ6bSaP look up an agent by name, id or wallet, and post what it owns, owes and has done. Sources on every figure
Read the finances behind every agent