Maple-Preview
- Input
- —
- Output
- —
- Speed
- ~120tok/sest.
- Context
- 131Ktokens
Rankings
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× H200 141GB
- native ternary weights
- Throughput
- ~16,365 tok/s
- many requests batched
- Price
- $0.023–0.19 /M tok
- busy → light use
- On your own machine
- RTX 5090 32GB (native ternary, 300+ tok/s single-stream)
- single consumer GPU or Mac
Assumptions (5)
- NATIVE TERNARY weights (20.2B params, 1.49B active per token) + 25% KV-cache headroom ≈ 5 GB
- 1× H200 141GB at $3.59–$7.91/GPU-hour on-demand (2026-09-24)
- ~16,365 output tok/s aggregate at batch 256 (bandwidth-bound); MoE compute scales with active params
- Blended 3:1 input:output; prefill ~313,784 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
~#4 of 21 (estimated)~80.6%
- #1Claude Opus 596.0%
- #2Claude Sonnet 585.2%
- #3Beam 1.0 501B A23B81.6%
- ~#4Maple-PreviewEstimated80.6%
- #4Gemini 3.1 Pro80.6%
No published score — estimated from related benchmarks
SWE-bench Pro (Public)Estimated
~#8 of 24 (estimated)~66.1%
- #1Claude Fable 5.293.0%
- #2Claude Opus 5.589.9%
- #7Beam 1.0 501B A23B66.1%
- ~#8Maple-PreviewEstimated66.1%
- #8Hy4 preview65.7%
No published score — estimated from related benchmarks
DeepSWE v1.1Estimated
~#43 of 46 (estimated)~33.6%
- #1Gemini 4 Argon78.8%
- #2DeepSeek V4.1 Flash 101777.0%
- #42Nex-N2.5-Mini36.1%
- ~#43Maple-PreviewEstimated33.6%
- #43Hy328.0%
No published score — estimated from related benchmarks
Terminal-Bench 4.0Estimated
~#24 of 102 (estimated)~31.6%
- #1Kimi K3.1 Pro67.2%
- #2Claude Sonnet 5.563.6%
- #22GLM-5.3 Flash32.8%
- ~#24Maple-PreviewEstimated31.6%
- #24DeepSeek V4.1 Flash26.8%
No published score — estimated from related benchmarks
LiveCodeBenchLab-reported
#14 of 2775.1%
- #1Qwen3.8-Omni-Flash92.6%
- #2Qwen3.8 Flash-Next91.9%
- #13Bonsai 27B76.4%
- #14Maple-Preview75.1%
- #15Mistral Small 464.0%
huggingface.co · 2026-09-30
SciCodeEstimated
~#36 of 119 (estimated)~53.1%
- #1Claude Fable 5.267.7%
- #2Claude Opus 5.566.9%
- #35MiniMax-M3.1-Flash53.4%
- ~#36Maple-PreviewEstimated53.1%
- #36Grok 4.653.0%
No published score — estimated from related benchmarks
AA Coding Agent IndexEstimated
~#11 of 26 (estimated)~51.0
- #1Claude Opus 5.561.7
- #2Claude Opus 559.7
- #10Muse Wonder52.0
- ~#11Maple-PreviewEstimated51.0
- #11GLM-5.348.9
No published score — estimated from related benchmarks
Hard reasoning
GPQA DiamondLab-reported
#102 of 19273.5%
- #1GPT-6 Astra96.1%
- #2Gemini 3.8 Flash95.3%
- #101Qwen3 Coder Next73.7%
- #102Maple-Preview73.5%
- #103Ling 3.0 Tiny73.4%
huggingface.co · 2026-09-30
Humanity's Last ExamEstimated
~#79 of 199 (estimated)~21.0%
- #1Claude Fable 5.274.3%
- #2Claude Opus 5.561.4%
- #78Step 3.7 Flash21.4%
- ~#79Maple-PreviewEstimated21.0%
- #79Nemotron 3 Super 120B A12B20.8%
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%
- #14Nova 2.0 Pro Preview89.0%
- #15Maple-Preview87.5%
- #15Ternary Bonsai 27B87.5%
huggingface.co · 2026-09-30
AA Intelligence IndexEstimated
~#63 of 103 (estimated)~20.5
- #1Gemini 4 Carbon59.4
- #2Claude Opus 5.557.6
- #62K2 Horizon 7B20.6
- ~#63Maple-PreviewEstimated20.5
- #63Ling 3.0 Flash20.1
No published score — estimated from related benchmarks