Analysis: Ternary Bonsai 2 9B vs 27B price, context, and GGUF
Two models share a 262144 context. Output is $0.13 vs $0.50 per million tokens, and the official GGUF download counts stay apart.
On 2026-11-02, Ternary Bonsai 2 9B and 27B are two listings: one context length, two prices, and two official GGUF repos.
Check Ternary Bonsai 2 9B and 27B. OpenRouter ids are prism-ml/ternary-bonsai-2-9b and prism-ml/ternary-bonsai-2-27b. The list is https://openrouter.ai/api/v1/models. The Hub search is https://huggingface.co/api/models?search=Ternary-Bonsai-2. sitdraw · Guide.
Two OpenRouter price rows
| Field | 9B | 27B |
|---|---|---|
| Model id | prism-ml/ternary-bonsai-2-9b | prism-ml/ternary-bonsai-2-27b |
| created (UTC) | 2026-10-16 | 2026-09-18 |
| context_length | 262144 | 262144 |
| Provider context | 262144 | 262144 |
| Max completion | 65536 | 32768 |
| Input / 1,000,000 tokens | $0.00000008 → $0.08 | $0.000000075 → $0.075 |
| Output / 1,000,000 tokens | $0.00000013 → $0.13 | $0.0000005 → $0.50 |
| Cache read / 1,000,000 tokens | $0.04 | $0.0375 |
| Modality | text+image → text | text+image → text |
| hugging_face_id | prism-ml/Ternary-Bonsai-2-9B-gguf | prism-ml/Ternary-Bonsai-2-27B-gguf |
| Reasoning | Optional, on by default; no effort list in the object | Optional, on by default; efforts xhigh / medium / none, default xhigh |
The price gap is on output
Both context_length values are 262144, so the window length does not fold them into one model. The split is the output price and the max completion. 27B input is $0.005 less per million tokens, and output is $0.37 more. The 9B description says it is built on Qwen3.5 with ternary weights packed into 2 bits. The 27B description says it is derived from Qwen3.8-27B. Both sentences come from that listing text. The cut-off tail is left unfilled.
Two official GGUF download counts
The Hugging Face search hits many repos. This table keeps only the two official cards named by OpenRouter hugging_face_id. 2B, 4B, MLX, and third-party GGUF files stay outside these cells.
| Field | 9B official GGUF | 27B official GGUF |
|---|---|---|
| modelId | prism-ml/Ternary-Bonsai-2-9B-gguf | prism-ml/Ternary-Bonsai-2-27B-gguf |
| downloads | 592665 | 1851236 |
| likes | 145 | 2885 |
| createdAt | 2026-09-23 | 2026-09-16 |
| pipeline_tag | image-text-to-text | text-generation |
Check it again yourself
- 01Open the OpenRouter list
Request
https://openrouter.ai/api/v1/modelsand find each id on its own. Readpricing.promptandpricing.completion, then multiply by 1000000 for USD per million tokens. - 02Match the Hub repos by id
Request
https://huggingface.co/api/models?search=Ternary-Bonsai-2. Keep onlyprism-ml/Ternary-Bonsai-2-9B-ggufandprism-ml/Ternary-Bonsai-2-27B-gguf. - 03Leave the two rows apart
A matching context only means that window field matches. Output price, max completion, created, and downloads each keep their own cell. Empty expiration_date and knowledge_cutoff stay empty.
Are 9B and 27B two names for one model?
They are two OpenRouter ids and two prism-ml GGUF repos. Parameter size, listing date, and output price each differ.
Does the same context mean the same bill?
The window is 262144 on both. Input per million tokens is $0.08 and $0.075. Output is $0.13 and $0.50. Max completion is 65536 and 32768.
Do 1851236 downloads mean 27B is better?
They mean the official 27B GGUF had that many downloads on 2026-11-02. The official 9B GGUF had 592665. Downloads and likes are not a quality score.
Why do the OpenRouter and Hub dates differ?
They are fields from two catalogs. 9B is 2026-10-16 on OpenRouter and 2026-09-23 on Hub createdAt. 27B is 2026-09-18 on OpenRouter and 2026-09-16 on Hub createdAt.
Two prices on one page
Spread this table, then leave when it is enough. sitdraw.