DeepGrove logo

Maple-Preview

MaplevPreviewCurrent
byDeepGroveDeepGrove(ai lab)
Released August 8, 2026
Context131.1K tokens
CategoryLarge Language Model

About Maple-Preview

Maple-Preview is DeepGrove’s open-source 20B-A1B ternary-weight reasoning LLM for efficient on-device inference. The model card reports 20.2B total parameters, 1.49B active parameters, a 131,072-token context window, and 5.31 GB checkpoint size. The model card specifies 131,072 tokens of context but does not establish a separate maximum output-token limit.

Capabilities

textreasoningcodeon device inference

Input Modalities

text

Output Modalities

text

Technical Details

API Identifier
deepgrove/maple-preview
Category
Large Language Model
Context Window
131,072 tokens

Tags

llmreasoning-modelon-device-aiternary-weightslow-precisionmoe

Benchmarks

Performance scores for Maple-Preview across standard benchmarks.

LiveCodeBench v6DeepGrove model card · Aug 2026
75.1
AIME 2026DeepGrove model card · Aug 2026
87.5
HMMT 2026DeepGrove model card · Aug 2026
78.8
GPQA DiamondDeepGrove model card · Aug 2026
73.5

Pricing and access

Open-source model card; no hosted API pricing found in the official source. Local inference costs depend on user hardware and runtime.

DeepGrove logo
DeepGroveai lab

Efficient low-precision language models for devices

Founded 2026Unknown
View full profile