1-bit and 1.58-bit LLMs that deliver full-size intelligence on edge devices
AI📍 Pasadena, CAFounded 2025
Current Valuation
Private
PRIVATE
IPO Outlookprivate
Is PrismML going public?
No. PrismML is private and has not filed to go public. There is no S-1 on file.
What is PrismML's valuation in 2026?
No confirmed valuation is on record for PrismML. The roster shows funding and investors where they are public; it does not estimate a mark.
Has PrismML filed an S-1?
No public S-1 is on file for PrismML.
Who are PrismML's investors?
PrismML's investors include Khosla Ventures, Cerberus Ventures, Caltech, Google (compute grant), Samsung (continuing support).
Answers are derived from the roster record shown on this page — filing status, last round and confirmed marks. Where the record is silent, the answer says so rather than guessing.
PrismML is a Pasadena, CA edge-AI company spun out of Caltech that builds extremely compressed large language models which run on phones, laptops, and embedded devices without sacrificing performance. It emerged from stealth on March 31, 2026; press coverage (WSJ CIO Journal, HPCwire) put the raise at $16.25M in a SAFE and seed round backed by Khosla Ventures, Cerberus Ventures, and Caltech, plus compute grants from Google and Caltech — the company's own announcement stated no figure, and no valuation has ever been publicly disclosed. Caltech owns the underlying patents and licenses them exclusively to PrismML. The 2026 product cadence ran from 1-bit Bonsai 8B (Mar, ~16GB compressed to ~1.15GB, ~8x faster inference, up to 80% less energy, ~44 tokens/s on an iPhone) through the 1.58-bit Ternary Bonsai family (Apr), Bonsai Image 4B and the Bonsai Studio iOS app (May), to flagship Bonsai 27B (Jul) — a Qwen3.6-27B compression at 5.9GB ternary / 3.9GB 1-bit with 262K context, which the company calls the first model of its capability class to run on a phone. All models ship free under Apache 2.0. In July 2026 CNBC reported Apple was evaluating the technology for on-device AI, with CEO Babak Hassibi confirming on the record that talks were early and progressing; Apple did not comment, and no agreement or design win exists. Performance claims — including 10-15x memory reduction — originate with PrismML and have not been independently validated. The team is led by Hassibi (a sitting Caltech professor) with co-founders Sahin Lale and Omead Pooladzandi as co-heads of research and Reza Sadri as VP of Strategy, plus CFO Shayan Ilbagian (ex-Cerberus).
Company Profile
CEO
Babak Hassibi
Founded
2025
HQ
Pasadena, CA
Employees
11-50
Total Raised
$16.25M
Growth
Pre-commercial — models open-sourced under Apache 2.0, free developer-preview API; no revenue disclosed
No verified Hugging Face organization found for PrismML — no attributable open-model footprint. Closed-weights labs are common; this is a distribution signal, not a capability judgment.
Checked huggingface.co as of 2026-08-20. Re-checked weekly.
Talent Signal
Hiring trend
● First snapshot
trend accrues with monthly sweeps
Open roles
8
via careers page
Product-build share
100%
engineering, research, product, design
Trajectory
steady
heavy product build — 100% of open roles are product-building functions (engineering/research/product) — 8 of 8 open roles are data.
Counted from PrismML's live job board as of 2026-08-17; role functions classified from posting titles. Open-role counts are an activity signal, not headcount. Refreshed monthly.
Private-company numbers are not real-time. Reflects publicly disclosed valuations from press releases, news reports, and tender offers as of recent date. Refreshed quarterly.