SHELFCRITTER AI Digest

MODEL

Bonsai 2 27B

modeltopic-noteedge-aiquantizationon-device

Overview

Bonsai 2 27B is PrismML‘s ternary-quantised ({−1, 0, +1}) derivative of Qwen 3.8 27B, launched Sept 17, 2026 under Apache 2.0. The model fits in 5.9 GB, carries a 262K context, and retains ~98.2% of the Qwen 3.8 27B baseline on an aggregate 20-benchmark score of 83.9. Distinct from Bonsai 27B‘s July 2026 quantisation of Qwen3.6-27B: Bonsai 2 targets the newer Qwen 3.8 base and pushes to a smaller ternary footprint. The framing that matters is that the baseline being matched is the same-size 27B model, compressed to phone-deployable footprint at ~98% quality retention, not a headline-friendly cross-size collapse. PrismML also disclosed a $22.25M seed round (Khosla, Cerberus, Google, with Samsung backing) in the same news cycle.

Timeline

  • 2026-09-18-AI-Digest — PrismML launches Bonsai 2 27B — ternary-quantised derivative of Qwen 3.8 27B, 5.9 GB / 262K context / Apache 2.0, aggregate ~98.2% of the Qwen 3.8 27B baseline (20-benchmark suite, 83.9). PrismML also discloses a $22.25M seed (Khosla, Cerberus, Google, with Samsung backing). Benchmarks are PrismML-reported — treat as vendor-claim until an independent bench (Aider, LM Arena) posts numbers. HN traction 326 pts / 106 cmts.

Key Developments

  1. Ternary compression of Qwen 3.8 27B to phone-deployable 5.9 GB (Sept 17, 2026): same-size baseline match at ~98.2%, not a cross-size scaling refutation. Apache 2.0. 262K context. 20-benchmark aggregate 83.9.
  2. PrismML $22.25M seed disclosed in the same news cycle (Khosla, Cerberus, Google, with Samsung backing) — not previously surfaced in the corpus.

Context