Ternary Bonsai 2 27B

Overview

Ternary Bonsai 2 27B is a 27-billion-parameter language model from Prism ML, released on September 18, 2026 and made available through OpenRouter. Its headline differentiator is the combination of a mid-sized 27B parameter scale with an unusually large 262,144-token context window. That places it in the category of long-context general-purpose models designed to process very large documents, codebases, transcripts, technical specifications, and multi-step reasoning tasks without aggressive chunking.

The “Ternary” name suggests an emphasis on highly compressed or low-precision model representation, although Prism ML has not publicly detailed the full training recipe, architecture, or quantization approach in the discovery data. In practical terms, the model is notable because it aims to offer long-context reasoning and generation in a smaller footprint than many frontier-scale systems.

Capabilities and features

Ternary Bonsai 2 27B supports text generation, reasoning, code generation, and long-context analysis. Its 262k-token context window is the standout feature: users can provide entire repositories, lengthy legal or policy documents, research papers, product manuals, meeting histories, or multi-file software projects in a single prompt. This can reduce retrieval overhead and improve continuity when the task depends on details scattered across a large input.

For reasoning tasks, the model is suited to structured analysis, comparison, summarization, planning, and explanation. For coding, it can generate functions, refactor snippets, explain unfamiliar code, assist with tests, and reason over broad project context when sufficient source files are included. Its value is likely highest in workflows where context breadth matters as much as raw benchmark performance.

Technical specifications

• Provider: Prism ML
• Model name: Ternary Bonsai 2 27B
• Parameters: 27B
• Context window: 262,144 tokens
• Modalities: Text input and text output, based on available listing data
• Primary capabilities: Text generation, reasoning, code generation, long-context processing
• Availability: Listed on OpenRouter
• Release date: 2026-09-18
• Max output tokens: Not specified in the available discovery data
• Pricing: Not specified in the provided listing data; users should check OpenRouter for current rates
• Open weights / license: Not confirmed from the available information

Strengths and benefits

The main strength of Ternary Bonsai 2 27B is its long-context capacity at a 27B scale. This makes it attractive for teams that need broad document awareness but may not require, or want to pay for, the largest frontier models. The model’s size may also make it more cost-efficient and responsive than larger systems, depending on Prism ML’s serving implementation and OpenRouter pricing.

Its context length is especially useful for tasks that suffer when information is split into fragments: whole-document summarization, cross-document comparison, codebase-level review, compliance analysis, and long conversational memory. If its ternary design reflects efficient inference, it may also offer favorable throughput or deployment economics, though that should be validated with real usage.

Limitations and caveats

The public information available for Ternary Bonsai 2 27B is limited. There are no confirmed benchmark scores, training data disclosures, safety evaluations, license terms, maximum output length, or pricing details in the provided discovery data. As with any long-context model, a 262k-token window does not guarantee perfect recall across the entire prompt; models can still miss details, overweight recent content, or produce plausible but incorrect conclusions.

Compared with larger frontier alternatives, a 27B model may be less capable on the hardest reasoning, advanced mathematics, complex agentic coding, or highly specialized expert tasks. Compared with smaller open models, however, it offers far more context capacity and likely stronger general performance. Against Prism ML’s prior Bonsai-family models, its primary advance appears to be the expanded scale and long-context capability, though direct comparisons require benchmark data.

For software maintenance, the large context window can be useful for reviewing dependency manifests, changelogs, lockfiles, and migration notes together, but that is best viewed as one application of its broader long-document reasoning capability.