Atlas

Models

← All models

Gemini 1.5 Flash-8B

Google DeepMind · 2024-10-03

Gemini 1.5 Flash-8B is Google's smallest Gemini API model; the exact stable gemini-1.5-flash-8b-001 snapshot became generally available on October 3, 2024 after experimental Flash-8B versions appeared in August and September. This multimodal transformer decoder is designed for high-throughput, low-latency uses such as chat, transcription, and long-context translation; the hosted model accepts text, images, audio, and video, returns text, and supports up to 1,048,576 input tokens. Google describes Flash-8B as a single-digit-billion-parameter model but does not disclose its exact total or active parameter count.

Benchmark scores

BenchmarkScore
AidanBench544
Artificial Analysis Intelligence Index5.5
BBH (BIG-Bench Hard)69.5
Capability113.8
Humanity's Last Exam (Text-Only)4.5
LisanBench Difficulty-Weighted24.4
LisanBench Path Length77.0
LiveCodeBench (Artificial Analysis)21.7
MATH35.9
MATH-500 (Artificial Analysis source)68.9
MGSM70.5
MMLU68.1
MMLU Pro56.9
MMMU50.3
MMMU Pro36.5
OTIS Mock AIME 2024-20254.6
SciCode (Artificial Analysis)22.9
Loading Atlas data…