prism-ml
PrismML: Ternary Bonsai 2 9B
prism-ml/ternary-bonsai-2-9b
Our take
AI-assisted analysis based on the sources below. Suitability is an editorial assessment; test it on your own workload.
Ternary Bonsai 2 9B is a compact open-weight vision-language model from PrismML, derived from Qwen3.5-9B with its architecture unchanged. Its roughly 8.95B-parameter language model uses ternary weights with FP16 group-wise scales, packed in GGUF formats of about 2 to 2.6 GB, plus an optional ~0.46B vision tower loaded only for image input. It inherits a 262,144-token context and runs via llama.cpp on CUDA, Metal and CPU, with an MLX variant for Apple Silicon. Documentation here is vendor-provided; no independent benchmarks were in this source pack.[1][2]
Strengths
- Extremely small footprint: the vendor reports ternary weights packed to about 2.0-2.56 GB, down from roughly 17.9 GB at FP16, enabling on-device use.[2]
- Long 262,144-token context inherited from the Qwen3.5-9B hybrid-attention backbone, per the catalog and model documentation.[1][2]
- Multimodal text-plus-image input via an optional vision tower, with tool-calling and reasoning parameters listed in the OpenRouter catalog.[1][2]
Limitations
- No independent benchmark results appear in this source pack; capability claims are vendor- or catalog-reported and should be validated on your actual workload.[1][2]
- Portability constraints: PTQ1_0 and PQ2_0 require the PrismML llama.cpp fork; upstream llama.cpp runs only the Q2_0 file from release b11556 onward, and older builds produce wrong output without warning.[2]
- Documented operational quirks: the BF16 mmproj can crash llama-server under concurrent image requests (Q8_0 is recommended), thinking mode can consume 3,000+ tokens, and greedy decoding may stall with thinking on.[2]
Where to use it
- On-device or memory-constrained chat assistants where a ~2 GB model footprint matters; worth piloting with quality checks on your data.[2]
- Image and document understanding trials, such as describing images or reading dense small text, using the optional mmproj vision pack.[1][2]
- Copy-heavy tasks like code edits and rewrites, where the vendor says the optional DFlash2 speculative drafter helps; MTP is suggested as the default.[2]
Sources
- [1] OpenRouter catalog · PrismML: Ternary Bonsai 2 9BSources checked: Nov 13, 2026
- [2] Model documentation · PrismML: Ternary Bonsai 2 9BSources checked: Nov 13, 2026
About this model
Description below is the original text supplied by OpenRouter, shown in the source language.
Ternary Bonsai 2 9B is PrismML’s compact, open-weight language model for conversational AI, reasoning, and tool use. Built on Qwen3.5, it uses ternary weights packed into 2 bits for efficient...
Specifications
- Model ID
- prism-ml/ternary-bonsai-2-9b
- Canonical ID
- prism-ml/ternary-bonsai-2-9b-20261015
- Creator
- prism-ml
- Added to OpenRouter
- 16.10.2026
- Context length
- 262,144
- Modality
- text+image->text
- Input modalities
- text, image
- Output modalities
- text
- Token pricing
- USD per 1M tokens: in $0.080 / out $0.130
- Knowledge cutoff
- —
- Expiration date
- —
Tasks and capabilities
A grade appears only where the source supplies a matching measurement for that task, the model is active and the cohort holds at least 10 comparable models. The capability column lists facts the source declares — supported formats, reasoning and tool parameters — which are not measured quality.
| Task | Grade | Source-declared capability |
|---|---|---|
| General quality | No measurement | reasoning: optional · parameters: include_reasoning, reasoning |
| Coding | No measurement | — |
| Agents | No measurement | parameters: structured_outputs, tool_choice, tools |
| Writing | No measurement | output: text |
| Reasoning | No measurement | reasoning: optional · parameters: include_reasoning, reasoning |
| Document analysis | No measurement | — |
| Vision | No measurement | input: image |
| Image generation | No measurement | — |
| Audio | No measurement | — |
| Video | No measurement | — |
| Embeddings | No measurement | — |
Evaluations
No evaluations are supplied for this model.
How grades work
Each letter grade compares a model only against other models scored by the same source and metric in the current snapshot. S means the top 10%, A the next 20%, B the next 30%, C the rest. Models without a published measurement are shown as not rated — a supported format alone never implies measured quality.
- S — S — top 10% of the cohort
- A — A — next 20% of the cohort
- B — B — next 30% of the cohort
- C — C — below the B band
Providers
Optional live provider statistics from OpenRouter. Missing values are unknown, not zero.
| Provider | Context length | Price in /1M | Price out /1M | Latency p50 (30m, s) | Throughput p50 (30m, tok/s) | Uptime (30m) |
|---|---|---|---|---|---|---|
| Darkbloomint4 | 262,144 | $0.080 | $0.130 | — | — | 100% |
Provider data retrieved: 13.11.2026
Related articles
No published articles identify this model yet.