PrismML: Ternary Bonsai 2 27B vs Sakana: Fugu Max
Side-by-side comparison of specs, pricing, benchmark scores, and task rankings. Updated 2026-10-13.
| PrismML: Ternary Bonsai 2 27B | Sakana: Fugu Max | |
|---|---|---|
| Vendor | prism-ml | sakana |
| Quality Score | 100 | 100 |
| Input Price | $0.07/M | $2.00/M |
| Output Price | $0.50/M | $6.00/M |
| Context Window | 262,144 | 1,000,000 |
| Max Output | 32,768 | 128,000 |
| Tool Calling | ✓ | ✓ |
| Structured Output | ✓ | ✓ |
| Reasoning Mode | ✓ | ✓ |
| Vision | ✓ | ✓ |
| Audio | - | - |
The verdict
For a sample job of 1 million input tokens and 250,000 output tokens, Ternary Bonsai 2 27B costs $0.20 and Fugu Max costs $3.50, so Ternary Bonsai 2 27B is about 17.5 times cheaper for the same work.
Fugu Max reads up to 1,000,000 tokens at once, roughly 1,500 pages of text, against about 393 pages for Ternary Bonsai 2 27B.
Across the 18 tasks where they differ, Ternary Bonsai 2 27B scores higher on 5 (Code Completion, Regex Writing, JSON Extraction and others), Fugu Max on 13 (SQL Generation, Code Review, Code Refactoring and others).
Pick Ternary Bonsai 2 27B if you need lower cost; pick Fugu Max if you need long documents or the stronger all-round task scores.
Costs use current list prices per million tokens; page counts assume about 0.75 words per token and 500 words per page.
Who wins by task?
| Task | PrismML: Ternary Bonsai 2 27B | Sakana: Fugu Max |
|---|---|---|
| SQL Generation | 131 | 133 |
| Code Review | 126 | 132 |
| Code Completion | 129 | 118 |
| Code Refactoring | 127 | 136 |
| Bug Fixing | 130 | 136 |
| Unit Test Generation | 121 | 124 |
| Code Documentation | 126 | 130 |
| Regex Writing | 119 | 118 |
| CI/CD Pipelines | 117 | 120 |
| CSV / Spreadsheet Cleanup | 128 | 133 |
| ETL Scripting | 122 | 128 |
| JSON Extraction | 131 | 122 |
| Bulk Data Labeling | 129 | 119 |
| OCR / Document Parsing | 128 | 131 |
| Table Extraction from PDFs | 128 | 131 |
| Long-Document Summarization | 129 | 137 |
| Short-Form Summarization | 123 | 114 |
| Blog Post Writing | 119 | 121 |
Scores reflect capability match + benchmark data + pricing for each task. Methodology →