---
type: "post"
id: "prismml-introducing-ternary-bonsai-2-2b-4b-and-8cd13c"
slug: "prismml-introducing-ternary-bonsai-2-2b-4b-and-8cd13c"
title: "PrismML — Introducing Ternary Bonsai 2 2B, 4B, and 9B: Bringing Local Agents to Everyday Devices"
description: "Thesis: Compact model releases make local agent deployment a memory-and-power allocation decision, but published capability retention varies by task."
retrieval_nugget: "Thesis: Compact model releases make local agent deployment a memory-and-power allocation decision, but published capability retention varies by task."
published_at: "2026-10-16"
updated_at: "2026-10-16"
record_date: "2026-10-16"
date_kind: "published_at"
topics: ["local-models","agents","models"]
source_urls: ["https://prismml.com/news/bonsai-2-family"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:prismml-introducing-ternary-bonsai-2-2b-4b-and-8cd13c"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:38be3f2748a20ea952e03c14d72742cf37beaef2520762bac2645ddf31af946d","reviewed_at":"2026-10-16T12:08:14.601Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":2,"watch_condition_count":1,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"Compact model releases make local agent deployment a memory-and-power allocation decision, but published capability retention varies by task.","observed_facts":[{"text":"PrismML announced Bonsai 2 models at 2B, 4B, and 9B parameters with reported footprints of 0.43GB, 0.95GB, and 2GB.","source_urls":["https://prismml.com/news/bonsai-2-family"]},{"text":"PrismML reports the 9B model retains 85.7% of Qwen3.5 9B's Tau2Bench score and 83.2% of its Omni_doc score.","source_urls":["https://prismml.com/news/bonsai-2-family"]},{"text":"PrismML lists NVIDIA GPU support through llama.cpp and Apple-device support through MLX, with weights under Apache 2.0.","source_urls":["https://prismml.com/news/bonsai-2-family"]}],"mechanism":"Ternary weights restricted to −1, 0, and +1 reduce model storage, allowing smaller deployments within device memory budgets. Smaller storage can enable on-device inference, but the benchmark results show that compression does not preserve every capability equally.","why_now":"PrismML has released downloadable weights under Apache 2.0 and describes support for common NVIDIA and Apple deployment paths, making the tradeoff actionable for developers evaluating local assistants.","implications":["Teams can test whether the memory and latency benefits justify task-specific quality loss before moving sensitive or intermittent workloads on-device.","Local execution may reduce dependence on network access, but the announcement alone does not establish privacy, total energy use, or production reliability."],"evidence_boundary":"All performance figures and comparisons here are PrismML-reported; the public announcement does not provide independent validation, and results may vary by hardware, workload, and evaluation setup.","watch_conditions":["Independent replication across devices and representative tool-use workflows would show whether the claimed footprint and speed translate into reliable local agents."],"related_records":[],"new_branch_reason":"No prior New Runtime analytical record was supplied for this specific PrismML release; this starts a distinct record on compact ternary models and local agent deployment tradeoffs."}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/prismml-introducing-ternary-bonsai-2-2b-4b-and-8cd13c/","markdown":"https://newruntime.com/posts/prismml-introducing-ternary-bonsai-2-2b-4b-and-8cd13c.md","json":"https://newruntime.com/posts/prismml-introducing-ternary-bonsai-2-2b-4b-and-8cd13c.json"}
---

# PrismML — Introducing Ternary Bonsai 2 2B, 4B, and 9B: Bringing Local Agents to Everyday Devices

## Retrieval answer

Thesis: Compact model releases make local agent deployment a memory-and-power allocation decision, but published capability retention varies by task.

Thesis: Compact model releases make local agent deployment a memory-and-power allocation decision, but published capability retention varies by task.

Observed facts: PrismML announced Bonsai 2 models at 2B, 4B, and 9B parameters with reported footprints of 0.43GB, 0.95GB, and 2GB. Observed facts: PrismML reports the 9B model retains 85.7% of Qwen3.5 9B's Tau2Bench score and 83.2% of its Omni_doc score. It lists NVIDIA GPU support through llama.cpp and Apple-device support through MLX.

Mechanism: Ternary weights restricted to −1, 0, and +1 reduce model storage, allowing smaller deployments within device memory budgets. Smaller storage can enable on-device inference, but the benchmark results show that compression does not preserve every capability equally.

Why now: PrismML has released downloadable weights under Apache 2.0 and describes support for common NVIDIA and Apple deployment paths, making the tradeoff actionable for developers evaluating local assistants.

Implications: Teams can test whether the memory and latency benefits justify task-specific quality loss before moving sensitive or intermittent workloads on-device. Implications: Local execution may reduce dependence on network access, but the announcement alone does not establish privacy, total energy use, or production reliability.

Evidence boundary: All performance figures and comparisons here are PrismML-reported; the public announcement does not provide independent validation, and results may vary by hardware, workload, and evaluation setup.

Watch conditions: Independent replication across devices and representative tool-use workflows would show whether the claimed footprint and speed translate into reliable local agents.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Local Models](https://newruntime.com/topics/local-models.md): Explore the local-models topic hub.
- topic hub: [Agents](https://newruntime.com/topics/agents.md): Explore the agents topic hub.
- topic hub: [Models](https://newruntime.com/topics/models.md): Explore the models topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)