Model

Gemini 1.0 Nano

CLOSED

Google · Gemini 1.0 family · released Dec 6, 2023

On-device model for Pixel 8 Pro (Nano-2, 3.25B variant); distilled and 4-bit quantized from larger Gemini models.

ReasoningCodingVisionFunction callingTool useAgentic

91.9

Elo · rank #387

Parameters

3.25B

Active params

3.25B (dense)

Context

32K tokens

Architecture

Dense Transformer decoder (distilled, 4-bit quantized for on-device)

License

Proprietary (on-device, Android AICore)

Languages

—

API price (in/out)

No hosted API

Modalities

text · vision

Benchmark results

Bar shows position within the tracked field; marker = field best

AI2DVision51.0%#41

best: Molmo 72B · 96.3%

BIG-Bench HardReasoning42.4%#110

best: ERNIE 4.5 300B-A47B · 94.3%

ChartQAVision51.9%#40

best: MiniMax-VL-01 · 91.7%

DocVQAVision74.3%#42

best: Qwen2-VL-72B · 96.5%

MATH-500Math22.8%#178

best: GPT-5 · 99.4%

MathVistaVision30.6%#61

best: Seed 2.1 Pro · 90.7%

MBPPCoding27.2%#100

best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%

MMLUKnowledge55.8%#219

best: OpenAI o3 · 92.9%

MMMUVision32.6%#106

best: Claude Fable 5 · 89.3%

TextVQAVision65.9%#25

best: Molmo 2 8B · 85.7%

VQAv2Vision67.5%#18

best: Reka Edge (2026) · 88.4%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

—

Output / M tok

—

Gemini 1.0 family

Elo progression across releases

Gemini 1.0 NanoDec 202391.9 Gemini 1.0 ProDec 2023620.1 Gemini 1.0 UltraFeb 2024992.0

Resources

DOCSstorage.googleapis.com/deepmind-media/gemini/gemini_1_report.pdf

API price — · each benchmark row carries its own source badge (see methodology)