Ternary Bonsai 2 Ternary 9B

Use it locally on Apple Silicon to answer questions about images or handle text-only prompts.

Availability

Downloadable from Hugging Face

Pricing model

Not established

Deployment

MLX on Apple Silicon (Metal)

Privacy

Not established

Licence

Apache 2.0

• The model accepts vision input through a separate mmproj pack. ^[1]

  - **Vision input** through a separate mmproj pack (Q8_0, or BF16 for reference precision), loaded only when an image arrives.

• Three GGUF language-model packings are provided: PTQ1_0, PQ2_0, and Q2_0. ^[1]

  - **Three GGUF packings** of the same weights, consumed directly by custom kernels and never expanded back to FP16: **PTQ1_0** (dense trits, 2.00 GB), **PQ2_0** (2-bit slots, 2.42 GB) and **Q2_0** (2.56 GB).

• The PrismML llama.cpp fork runs all listed files on CUDA, Metal, and CPU.
• The model supports vision input through its official vision tower. ^[2]

  - **Vision input** through the official vision tower, carried unrotated in FP16.

• The model has a 262K-token context. ^[2]

  - **262K-token context**, inherited from the Qwen3.5-9B hybrid-attention backbone (24 linear-attention and 8 full-attention layers).

• The model is available as a downloadable MLX pack on Hugging Face.
• The pack runs with MLX on Apple Silicon. ^[2]

  A 9B-class vision-language model with ternary transformer weights, packed for MLX on Apple Silicon.

• The model pack is 3.46 GB on disk. ^[2]

  > **3.46 GB** on disk, language model + vision tower (FP16 language model: ~17.9 GB)

• The optional trained multi-token-prediction head is intended to help most on copy-heavy work such as code edits and rewrites. ^[2]

  It helps most on copy-heavy work such as code edits and rewrites; on general chat it is about as fast as plain decoding or slightly slower, so plain decoding stays the default.

• Bespoke-Nimble-9B-v2 leads 12 AI developments on 15 October 2026 Daily brief 2026-10-15