PrismML: Ternary Bonsai 2 27B

PrismML: Ternary Bonsai 2 27B is a text-focused model from PrismML, released by the provider on 18 September 2026. It is a 27B-parameter reasoning model derived from Qwen3.8-27B, and it sits in the text category on api14. If you are choosing a model for a text pipeline, the important context is that this one is built for reasoning work rather than general chat alone.

The model accepts text and images as input and returns text. Its context window is 262,144 tokens, and the maximum output is 32,768 tokens. According to the provider description, it supports coding, mathematics, tool calling, and image understanding. Ternary compression is part of how the model is built, which is worth noting if model size or deployment characteristics matter to you. Reasoning behaviour is exposed through request parameters rather than being fixed, so you can control how much reasoning the model performs.

Those capabilities point to a few practical uses. Because it handles code and mathematics, it fits tasks such as code generation, code explanation, and step-by-step problem solving where the reasoning trace matters. Image understanding means it can also take screenshots, diagrams, or other images alongside a text prompt, which is useful for tasks like reading a chart or interpreting a UI screenshot. Tool calling and structured outputs make it a candidate for agents and pipelines that need the model to call functions or return data in a defined shape rather than free text.

You call it through api14's OpenAI-compatible API using the model id "prism-ml/ternary-bonsai-2-27b". The request parameters it supports are frequency_penalty, include_reasoning, logprobs, max_tokens, presence_penalty, reasoning, reasoning_effort, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_logprobs, and top_p. If you already use an OpenAI-style client, the main change is the model id and the parameters you set. Use max_tokens to cap output, since the maximum output is 32,768 tokens even though the context window is much larger.

There are limits worth planning around. Input is text and image, but output is text only, so it will not produce images or audio. If your workload is purely conversational and does not need reasoning, tool calling, or image input, a simpler text model may be a better fit. For long documents, the 262K-token context window is generous, but remember that the 32,768-token output cap applies separately. The provider description is the source for the coding, mathematics, tool calling, and image understanding claims, so treat those as the stated scope rather than measured performance.