Gemini 1.5 Flash: Complete Benchmark Performance Guide

google 1.5-flash 2024-05-14 Proprietary

In-depth analysis of Gemini 1.5 Flash performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendorgoogle
Version1.5-flash
Release Date2024-05-14
Context Window1e+06 tokens
Input Modalitiestext, image, audio, video
Output Modalitiestext
LicenseProprietary
Documentationhttps://ai.google.dev/gemini-api/docs

Benchmark Performance

BenchmarkScoreUnitEvaluated AtNotesSource
MMLU76%2024-05-145-shotview
HUMANEVAL67.5pass@12024-05-14—view
GSM8K75.2%2024-05-140-shot CoTview
MATH53.2%2024-05-140-shot CoTview
BBH71.7%2024-05-143-shot CoTview
GPQA38.5%2024-05-140-shotview
IFEVAL68%2024-05-14prompt_strictview
ARC93.4%2024-05-14challengeview
MUSR46.1%2024-05-140-shotview
WINOGRANDE78.9%2024-05-140-shotview

Pricing

TierPriceCurrency
Input$0.075 / MtokUSD
Output$0.3 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://ai.google.dev/pricing · as of 2024-05-14

Compliance

  • Data Residency: US
  • SOC2: ✓
  • HIPAA: ✗
  • GDPR: ✓
  • ISO 27001: ✓

Gemini 1.5 Flash

Descripción del modelo

Google Gemini 1.5 Flash 经济型模型, 1M 上下文, 速度快成本低, 适合大规模多模态任务。

Especificaciones principales

ProveedorVersiónFecha de lanzamientoVentana de contextoModalidades de entradaModalidades de salidaLicencia
Google1.5-flash2024-05-141000Ktext, image, audio, videotextProprietary

Rendimiento en benchmarks

BenchmarkPuntuaciónUnidadNotas
MMLU (Massive Multitask Language Understanding)76.0%5-shot
HumanEval67.5pass@1—
GSM8K (Grade School Math 8K)75.2%0-shot CoT
MATH53.2%0-shot CoT
BBH (BIG-Bench Hard)71.7%3-shot CoT
GPQA38.5%0-shot
IFEval68.0%prompt_strict
ARC93.4%challenge
MUSR46.1%0-shot
WinoGrande78.9%0-shot

Precios

EntradaSalidaLectura cachéEscritura caché
————

por millón de tokens

Fortalezas

  • 支持文本、图像、音频多模态输入。
  • 上下文窗口 1000K，支持长文本。

Debilidades

  • 闭源专有模型，不支持自托管。

Casos de uso

  • 长文档摘要
  • 视觉与图像理解

Referencias