Ternary Bonsai 2 27B

Ternary Bonsai 2 27B is a reasoning model from Prism ML in the Ternary Bonsai 2 family. 13 benchmarks count toward its score, in 7 categories.

availableShows if the model has enough results for an index.
IndexOverall score out of 100.52.7 ±5.0
CoverageShare of the index weight with results.95%
SpeedOutput tokens per second.10/s
Input / 1MUS dollars per 1M input tokens.$0.075
Output / 1MUS dollars per 1M output tokens.$0.5
ContextMaximum tokens in one request.262K
EloLMArena rating and rank.N/A

The index is a score out of 100. The ± range shows how much it can change.

CapabilitiesScore per category, out of 100.

Out of 100
AgenticMulti-step tasks with tools.
55.8
CodingCode writing and repair.
52.7
ReasoningLogic problems and puzzles.
61.5
MultimodalTasks with images and text.
46.7
KnowledgeFacts and expert knowledge.
51.0
MultilingualTasks in many languages.
N/A
InstructionTasks with strict rules in the prompt.
50.1
MathMath problems.
52.8

Results

13 counted
BenchmarkThe test name.CategoryThe capability that the test measures.ResultThe score from the publisher.IndexThis result as a score out of 100.RunThe settings of the run.DateDate of the result.Published byThe source of the result.
MATH-500 Problem SetMath98.8%52.2Dan Hendrycks et al.
Grade School Math 8KMath96.7%DeepSeek-AI
AIME 2026Math95.8%54.8Qwen
American Invitational Mathematics Examination 2025Math95.0%51.4Mathematical Association of America
Instruction-Following EvalInstruction91.3%48.0Jeffrey Zhou et al.
LiveCodeBench v6Coding90.1%59.0LiveCodeBench maintainers
OmniDocBench v1.6Multimodal89.1%Linke Ouyang et al.
MMLU-ReduxKnowledge89.1%44.7Qwen
A Benchmark for Visual Question Answering using World KnowledgeMultimodal86.8%Dustin Schwenk et al.
Graduate-Level Google-Proof Q&AKnowledge85.8%57.4David Rein et al.
GPQA DiamondKnowledge85.8%57.4David Rein et al.
τ²-Bench Tool-Agent-User EvaluationAgentic80.2%56.4Victor Barres et al.
RealWorldQAMultimodal80.1%46.7Qwen
CharXiv Descriptive and Reasoning CombinedMultimodal80.0%CharXiv authors
Artificial Analysis Long Context ReasoningReasoning77.0%61.5Artificial Analysis
Berkeley Function Calling Leaderboard v3Agentic74.9%Shishir G. Patil et al.
Instruction Following BenchmarkInstruction74.0%52.2Benchmark authors
Software Engineering Benchmark VerifiedCoding60.8%46.3Carlos E. Jimenez et al.
BigCodeBenchCoding58.1%DeepSeek-AI
OCRBench V2Multimodal56.9%OCRBench authors
Terminal-Bench 2.1 (provider run)Agentic52.8%55.2DeepSeek-AI
Terminal-Bench 2.1 (provider run)Agentic52.8%55.2DeepSeek-AI

13 benchmarks count, from 15 of 22 results. A grey row does not count. Too few models took that benchmark.

Sources

BenchLM benchmark aggregationUsed with attribution; per-benchmark results credited to their original authorsOpenRouter, collected directlyNo licence stated

Same level, lower price

Qwen3.8-Flash-Next66.4 · Freedots3-note Preview65.5 · FreeApodex 1.1 Mini59.1 · Free

More from Prism ML

1-bit Bonsai 1.7BUnranked1-bit Bonsai 27BUnranked1-bit Bonsai 4BUnranked1-bit Bonsai 8BUnrankedTernary Bonsai 1.7BUnranked