What AI models can the Google Coral Dev Board (4 GB) run?

RAM: 4.0 GB
Usable for models: 2.8 GB
Memory bandwidth: 13 GB/s
Runs models on: CPU cores
Released: 2019

126 / 622 LLMs run Best: LFM3 1.7B 7 / 61 Image models run Best: Realistic Vision V5.1 1 / 43 Video models run Best: AnimateDiff v3 (SD 1.5) 41 / 70 Decision models run Best: Jina Reranker v3.5

Models run on the CPU cores here, so the memory’s speed and how fast those cores can read it set the pace: small models (about 1-4B) are usable, bigger ones are slow. We plan around 2.8 GB of the 4.0 GB (what Linux leaves free for a model). Edge TPU: 4 TOPS (INT8) for TensorFlow Lite models compiled for it. It has about 8 MB of on-chip memory, so it can’t hold a language model; the USB and M.2 Coral accelerators are the same Edge TPU.

More about this hardware

Maker	Google
Type	Single-board computer or edge AI board
Share of memory for models	70% (what Linux typically leaves free on a board this size)
Chip	NXP i.MX 8M
AI accelerator	Edge TPU: 4 TOPS (INT8) for TensorFlow Lite models compiled for it. It has about 8 MB of on-chip memory, so it can’t hold a language model; the USB and M.2 Coral accelerators are the same Edge TPU.
Boards with this chip	Coral Dev Board (4 GB version)
Memory type	LPDDR4, 32-bit, 1600 MHz clock
Bandwidth figure	12.8 GB/s, as the maker states it
Note	4× Cortex-A53 at 1.5 GHz. Language models run on the CPU only: tiny ones (under 1B) at a few tokens per second (estimate, at about 12% of the rated memory speed). The original 1 GB version leaves too little memory for language models and isn’t listed.
Specifications	coral.ai

Apps that run models well on the Google Coral Dev Board (4 GB)

Linux on ARM boards

For a typical model that fits in memory. Each model page picks the best app for that model on this hardware. All apps

Best LLMs for the Google Coral Dev Board (4 GB)

Quality	Model	Quant	Memory	Speed	Runs
55Good quality (B) · 55 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking.	LFM3 1.7BLiquid AI	`Q6_K`	2.3 GB	1.1 tok/s	Very slow
57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking.	MiniCPM5 2BOpenBMB	`Q4_K_M`	2.7 GB	1.0 tok/s	Very slowtight fit
48Fair quality (C) · 48 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking.	LFM3-VL 2BLiquid AI	`Q5_K_M`	2.3 GB	1.1 tok/s	Very slow
46Fair quality (C) · 46 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking.	Spark-X2.6 4BSparkLLM	`Q2_K`	2.8 GB	1.0 tok/s	Very slowtight fit
42Fair quality (C) · 42 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking.	TMax 4BAi2	`Q2_K`	2.8 GB	1.0 tok/s	Very slowtight fit
39Basic quality (D) · 39 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking.	MiniCPM-V 4.7 1BOpenBMB	`Q6_K`	2.2 GB	1.2 tok/s	Very slow
41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking.	LFM2.5 2.6BLiquid AI	`Q4_K_M`	2.6 GB	1.0 tok/s	Very slowtight fit
38Basic quality (D) · 38 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking.	TMax 2BAi2	`Q5_K_M`	2.3 GB	1.2 tok/s	Very slow
41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking.	Jamba Reasoning 3BAI21 Labs	`Q4_K_M`	2.8 GB	0.8 tok/s	Very slowtight fit
38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking.	Qwen3.5 2BAlibaba	`Q4_K_M`	2.3 GB	1.2 tok/s	Very slow
39Basic quality (D) · 39 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking.	Jamba2 3BAI21 Labs	`Q4_K_M`	2.7 GB	0.9 tok/s	Very slowtight fit
30Basic quality (D) · 30 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking.	Unlimited OCR 2 (3B-A0.6B)Baidu	`Q2_K`	2.7 GB	7.0 tok/s	Slowtight fit

Best Image models for the Google Coral Dev Board (4 GB)

Quality	Model	Quant	Memory	Speed	Runs
42Fair quality (C) · 42 of 100: how capable the model is at FP16, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking.	Realistic Vision V5.1SG161222	`FP16`	2.5 GB	8.7 min	Very slowtight fit · encoder unloaded after use
28Basic quality (D) · 28 of 100: how capable the model is at FP16, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking.	SD TurboStability AI	`FP16`	2.6 GB	35 s	Runs welltight fit · encoder unloaded after use
40Fair quality (C) · 40 of 100: how capable the model is at FP16, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking.	DreamShaper 8Lykon	`FP16`	2.5 GB	8.7 min	Very slowtight fit · encoder unloaded after use
50Fair quality (C) · 50 of 100: how capable the model is at FP16, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking.	PixArt-Σ XL 1024PixArt	`FP16`	2.3 GB	13.5 min	Very slowCPU offload
30Basic quality (D) · 30 of 100: how capable the model is at FP16, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking.	LCM DreamShaper v7Tsinghua (LCM)	`FP16`	2.5 GB	1.8 min	Slowtight fit · encoder unloaded after use
35Basic quality (D) · 35 of 100: how capable the model is at FP16, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking.	Stable Diffusion 1.5Stability AI / Runway	`FP16`	2.5 GB	8.7 min	Very slowtight fit · encoder unloaded after use

Best Video models for the Google Coral Dev Board (4 GB)

Quality	Model	Quant	Memory	Speed	Runs
35Basic quality (D) · 35 of 100: how capable the model is at FP16, whatever it runs on. Match score 42/100 (quality, fit, speed and memory use) sets the ranking.	AnimateDiff v3 (SD 1.5)Shanghai AI Lab	`FP16`	2.6 GB	3.0 h	Very slowtight fit · CPU offload

Best Decision models for the Google Coral Dev Board (4 GB)

Quality	Model	Quant	Memory	Speed	Runs
86Excellent quality (A+) · 86 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking.	Jina Reranker v3.5Jina AI	`Q8_0`	1.7 GB	51.7 s	Very slow
84Great quality (A) · 84 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking.	Jina Reranker v3Jina AI	`Q8_0`	1.7 GB	51.7 s	Very slow
80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking.	GLiNER2.5 DecideFastino	`Q8_0`	1.2 GB	22.7 s	Very slow
81Great quality (A) · 81 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking.	mxbai-rerank-large v2Mixedbread	`Q6_K`	2.1 GB	132.3 s	Very slow
80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking.	FRIDA-DecisionsSber AI	`Q8_0`	1.7 GB	38.0 s	Very slow
80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking.	Qwen3 Reranker 0.6BAlibaba	`Q8_0`	1.7 GB	51.7 s	Very slow
77Great quality (A) · 77 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking.	zerank-1 smallZeroEntropy	`Q5_K_M`	2.3 GB	147.6 s	Very slow
78Great quality (A) · 78 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking.	OpenAI Privacy FilterOpenAI	`Q8_0`	2.4 GB	2.3 s	Very slowtight fit

Best LLMs by task on this hardware Open in Model Finder

Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.