Ternary Bonsai 27B
- Input
- —
- Output
- —
- Speed
- ~84tok/sest.
- Context
- 262Ktokens
Rankings
Coding
Lab-reported#40 / 137rank by quality62.1±10.4quality / 1001/7 benchmarks measuredWriting & chat
Lab-reported#46 / 162rank by quality62.9±9.7quality / 1001/5 benchmarks measuredVision
Lab-reported#48 / 98rank by quality56.7±8.7quality / 1001/6 benchmarks measuredHard reasoning
Lab-reported#55 / 218rank by quality64.2±9.9quality / 1001/4 benchmarks measured
Self-hostingEstimated
How we estimate →What it would cost to run these open weights yourself on rented GPUs. No API sells this model, so this is its price on the chart.
- Hardware
- 1× B200 180GB
- ternary (Q2_0_g128) weights
- Throughput
- ~6,372 tok/s
- many requests batched
- Price
- $0.13–1.16 /M tok
- busy → light use
- On your own machine
- RTX 5090 32GB (ternary (Q2_0_g128), ~210 tok/s single-stream)
- single consumer GPU or Mac
Assumptions (5)
- TERNARY (Q2_0_G128) weights (27.3B params) + 25% KV-cache headroom ≈ 7 GB
- 1× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
- ~6,372 output tok/s aggregate at batch 182 (bandwidth-bound)
- Blended 3:1 input:output; prefill ~38,942 tok/s
- Low = 75% utilization at the low GPU price; high = 20% at the high price
Benchmark scores
~ italic, dashed = no published score yet, estimated from related benchmarks.
Coding
SWE-bench VerifiedEstimated
~#5 of 21 (estimated)~80.6%
- #1Claude Opus 596.0%
- #2Claude Sonnet 585.2%
- #4Gemini 3.1 Pro80.6%
- ~#5Ternary Bonsai 27BEstimated80.6%
- #5MiniMax-M380.5%
No published score — estimated from related benchmarks
SWE-bench Pro (Public)Estimated
~#9 of 24 (estimated)~65.6%
- #1Claude Fable 5.293.0%
- #2Claude Opus 5.589.9%
- #8Hy4 preview65.7%
- ~#9Ternary Bonsai 27BEstimated65.6%
- #9GPT-5.6 Terra63.4%
No published score — estimated from related benchmarks
DeepSWE v1.1Estimated
~#43 of 46 (estimated)~33.3%
- #1Gemini 4 Argon78.8%
- #2DeepSeek V4.1 Flash 101777.0%
- #42Nex-N2.5-Mini36.1%
- ~#43Ternary Bonsai 27BEstimated33.3%
- #43Hy328.0%
No published score — estimated from related benchmarks
Terminal-Bench 4.0Estimated
~#24 of 102 (estimated)~30.6%
- #1Kimi K3.1 Pro67.2%
- #2Claude Sonnet 5.563.6%
- #22GLM-5.3 Flash32.8%
- ~#24Ternary Bonsai 27BEstimated30.6%
- #24DeepSeek V4.1 Flash26.8%
No published score — estimated from related benchmarks
LiveCodeBenchLab-reported
#10 of 2782.8%
- #1Qwen3.8-Omni-Flash92.6%
- #2Qwen3.8 Flash-Next91.9%
- #9Nemotron 3.5 Super VL NVFP487.4%
- #10Ternary Bonsai 27B82.8%
- #11Gemma 4 31B80.0%
huggingface.co · 2026-09-30
SciCodeEstimated
~#36 of 119 (estimated)~53.2%
- #1Claude Fable 5.267.7%
- #2Claude Opus 5.566.9%
- #35MiniMax-M3.1-Flash53.4%
- ~#36Ternary Bonsai 27BEstimated53.2%
- #36Grok 4.653.0%
No published score — estimated from related benchmarks
AA Coding Agent IndexEstimated
~#11 of 26 (estimated)~50.2
- #1Claude Opus 5.561.7
- #2Claude Opus 559.7
- #10Muse Wonder52.0
- ~#11Ternary Bonsai 27BEstimated50.2
- #11GLM-5.348.9
No published score — estimated from related benchmarks
Writing & chat
IFBenchLab-reported
#59 of 13958.5%
- #1MiniMax-M382.9%
- #2Qwen3.8 2.4T A95B82.8%
- #58DiffusionGemma 26B A4B59.5%
- #59Ternary Bonsai 27B58.5%
- #60NVIDIA Nemotron 3 Nano 4B58.2%
huggingface.co · 2026-09-30
LMArena Text, Non-English (Elo)Estimated
~#28 of 57 (estimated)~1434 Elo
- #1Gemini 4 ArgonFew votes1512
- #2Claude Opus 5.51497
- #27GPT-6 Luna1436
- ~#28Ternary Bonsai 27BEstimated1434
- #28Qwen3.5 397B A17B1434
No published score — estimated from related benchmarks
LMArena Text (Elo)Estimated
~#27 of 57 (estimated)~1448 Elo
- #1Gemini 4 Argon1525
- #2Claude Opus 5.51507
- #26GPT-5.4 mini1448
- ~#27Ternary Bonsai 27BEstimated1448
- #27Step 5 Preview1447
No published score — estimated from related benchmarks
EQ-Bench Creative Writing v3 (Elo)Estimated
~#9 of 26 (estimated)~1955 Elo
- #1GPT-6 Astra2173
- #2Claude Fable 5.12162
- #8Grok 4.72007
- ~#9Ternary Bonsai 27BEstimated1955
- #9Muse Spark 1.31906
No published score — estimated from related benchmarks
Vision
MMMU-ProLab-reported
#50 of 8769.0%
- #1Claude Opus 5.587.7%
- #2GPT-6 Astra86.9%
- #48Qwen3.5 9B69.2%
- #50Ternary Bonsai 27B69.0%
- #51EXAONE 4.5 33B67.3%
huggingface.co · 2026-09-30
CharXiv ReasoningEstimated
~#9 of 12 (estimated)~78.4%
- #1Gemini 3.8 Flash86.2%
- #2Kimi K384.8%
- #8Muse Glimmer78.8%
- ~#9Ternary Bonsai 27BEstimated78.4%
- #9Inkling78.1%
No published score — estimated from related benchmarks
OmniDocBench v1.5Estimated
~#4 of 8 (estimated)~85.3%
- #1MiniMax-M391.6%
- #2Kimi K391.1%
- #2Qwen3.8 27B91.1%
- ~#4Ternary Bonsai 27BEstimated85.3%
- #4Gemini 3.1 Pro85.3%
No published score — estimated from related benchmarks
GDP.pdf (AA)Estimated
~#26 of 48 (estimated)~14.4%
- #1Muse Wonder Full32.6%
- #2GPT-6 Astra31.0%
- #25Step 5 Preview14.8%
- ~#26Ternary Bonsai 27BEstimated14.4%
- #26Gemini 3.5 Flash-Lite13.6%
No published score — estimated from related benchmarks
ChartographyEstimated
~#17 of 25 (estimated)~20.1%
- #1Claude Fable 5.273.4%
- #2GPT-6 Astra71.0%
- #16Grok 4.620.5%
- ~#17Ternary Bonsai 27BEstimated20.1%
- #17Muse Glimmer17.2%
No published score — estimated from related benchmarks
Hard reasoning
GPQA DiamondEstimated
~#65 of 193 (estimated)~83.6%
- #1GPT-6 Astra96.1%
- #2Gemini 3.8 Flash95.3%
- #64Gemini 3.5 Flash-Lite83.8%
- ~#65Ternary Bonsai 27BEstimated83.6%
- #65Muse Glimmer83.5%
No published score — estimated from related benchmarks
Humanity's Last ExamEstimated
~#56 of 199 (estimated)~31.2%
- #1Claude Fable 5.274.3%
- #2Claude Opus 5.561.4%
- #55Nemotron 3.5 Super VL NVFP431.4%
- ~#56Ternary Bonsai 27BEstimated31.2%
- #56Nemotron 3.5 Super VL30.7%
No published score — estimated from related benchmarks
AIME (latest)Lab-reported
#15 of 3387.5%
- #1GLM-5.3 NVFP499.5%
- #2Beam 1.0 501B A23B97.7%
- #15Maple-Preview87.5%
- #15Ternary Bonsai 27B87.5%
- #17Bonsai 27B87.1%
huggingface.co · 2026-09-30
AA Intelligence IndexEstimated
~#37 of 103 (estimated)~34.1
- #1Gemini 4 Carbon59.4
- #2Claude Opus 5.557.6
- #36DeepSeek V4 Pro (0813)36.0
- ~#37Ternary Bonsai 27BEstimated34.1
- #37Qwen3.8 27B33.7
No published score — estimated from related benchmarks