Sipsa Labs, Inc.

An experimental and deep tech-and-software company (Delaware C-corp, incorporated May 2026). We invent and ship across the full breadth of tech and software — deep research, runtime systems, novel substrates, infrastructure, hardware-adjacent stacks, and software products that don't fit anywhere else yet. UltraCompress — lossless 5-bit transformer compression — is our first flagship publicly-shipped product. More products in flight.
Live on Hacker News today: news.ycombinator.com/item?id=48099107
v0.6.2 on PyPI 22 architectures verified USPTO 64/049,511 + 64/049,517 BUSL-1.1 + Additional Use Grant

2026-05-11 — 22 architectures verified at 5 bpw, three new sub-1.005× records this week

Production-grade lossless 5-bit transformer compression across 22 architectures, dense + Mixture-of-Experts + state-space (Mamba), 0.6B to 405B parameters. Mathematically lossless customer-side reconstruction: SHA-256 over reconstructed tensor bytes matches the trainer's measurement, verified by uc verify.

Tightest verified PPL ratios at 5 bpw

(perplexity ratio = compressed PPL / bf16 baseline PPL; FineWeb-edu held-out tail; seq_len = 1024; n = 30-50; seed = 42)

ModelParamsTypePPL ratioNotes
Phi-3-mini-4k-instruct3.8Bdense1.00262×seq_len=128 caveat
Mixtral-8x7B47BMoE1.00368×tightest MoE result
Qwen3-1.7B-Base1.7Bdense1.00401×small-decoder record
Qwen3-14B14Bdense1.00403×14B-class record
Yi-1.5-9B8.8Bdense1.00414×>8B record
Qwen3-8B8Bdense1.00440×8B-class record
Mistral-7B-v0.37Bdense1.00548×NEW this week — 9.16× tighter than prior
Hermes-3-Llama-3.1-405B405Bdense1.0066×largest dense 5-bit lossless on the Hub
Qwen3-0.6B0.6Bdense1.0069×
OLMo-2-0425-1B1Bdense1.0073×
SmolLM2-1.7B-Instruct1.7Bdense1.0075×
Mamba-2.8B2.8BSSM1.0119×first published 5-bit lossless on a state-space model
Llama-3.1-8B8Bdense1.0125×standard eval

Source + reproduce: github.com/sipsalabs/ultracompress · full benchmark page: sipsalabs.com/inference

Quick start (3 commands, no GPU required for verify)

pip install ultracompress
hf download SipsaLabs/qwen3-8b-uc-v3-bpw5 --local-dir ./qwen3-8b
uc bench ./qwen3-8b

Inference API

api.sipsalabs.com is live, OpenAI-compatible:

export OPENAI_BASE_URL=https://api.sipsalabs.com/v1
# The official `openai` SDK works unchanged.

Pricing tiers + Compression-as-a-Service contracts at sipsalabs.com/pricing.


License + IP


Contact

sipsalabs.com · github.com/sipsalabs · @SipsaLabs on X

Public verifier dashboard: sipsalabs.com/inference · Selective Disclosure Charter — what we publish vs what we keep internal: see github.com/sipsalabs/ultracompress