Gemini 1.0 Flash：完整基准性能指南

google 1.0-flash 2024-02-15 Proprietary

深入分析 Gemini 1.0 Flash 在 MMLU、HumanEval、GSM8K、MATH、BBH 等基准上的表现，包含定价与适用场景。

Key Specifications

Vendorgoogle
Version1.0-flash
Release Date2024-02-15
Context Window32768 tokens
Input Modalitiestext, image
Output Modalitiestext
LicenseProprietary
Documentationhttps://ai.google.dev/gemini-api/docs

Benchmark Performance

BenchmarkScoreUnitEvaluated AtNotesSource
MMLU78.7%2024-02-155-shotview
HUMANEVAL65.7pass@12024-02-15—view
GSM8K86.2%2024-02-150-shot CoTview
MATH39.9%2024-02-150-shot CoTview
BBH74.8%2024-02-153-shot CoTview
GPQA37.7%2024-02-150-shotview
IFEVAL71.8%2024-02-15prompt_strictview
ARC92.6%2024-02-15challengeview
MUSR55.9%2024-02-150-shotview
WINOGRANDE80.1%2024-02-150-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 2024-02-15

Compliance

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

Gemini 1.0 Flash

模型概述

Google Gemini 1.0 Flash 经济型, 32K 上下文, 速度快成本低, 适合实时多模态应用。

核心规格

厂商版本发布日期上下文窗口输入模态输出模态许可证
Google1.0-flash2024-02-1532Ktext, imagetextProprietary

基准测试表现

基准得分单位备注
MMLU (Massive Multitask Language Understanding)78.7%5-shot
HumanEval65.7pass@1—
GSM8K (Grade School Math 8K)86.2%0-shot CoT
MATH39.9%0-shot CoT
BBH (BIG-Bench Hard)74.8%3-shot CoT
GPQA37.7%0-shot
IFEval71.8%prompt_strict
ARC92.6%challenge
MUSR55.9%0-shot
WinoGrande80.1%0-shot

定价

输入输出缓存读取缓存写入
————

每百万 token

优势

  • GSM8K 86.2，数学推理稳健。

劣势

  • 闭源专有模型，不支持自托管。

适用场景

  • 通用对话与问答

参考文献