Key Specifications

Vendorgoogle
Versiontext-bison
Release Date2023-05-10
Context Window8192 tokens
Input Modalitiestext
Output Modalitiestext
LicenseProprietary
Documentationhttps://ai.google.dev/gemini-api/docs

Benchmark Performance

BenchmarkScoreUnitEvaluated AtNotesSource
MMLU53.9%2023-05-105-shotview
HUMANEVAL45.1pass@12023-05-10view
GSM8K42.8%2023-05-100-shot CoTview
MATH12.1%2023-05-100-shot CoTview
BBH48.2%2023-05-103-shot CoTview
GPQA27.7%2023-05-100-shotview
IFEVAL59.5%2023-05-10prompt_strictview
ARC86.7%2023-05-10challengeview
MUSR40.8%2023-05-100-shotview
WINOGRANDE74.1%2023-05-100-shotview

Pricing

TierPriceCurrency
Input$0.5 / MtokUSD
Output$1 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://ai.google.dev/pricing · as of 2023-05-10

Compliance

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

Text Bison

模型概述

Google Vertex AI Text Bison 文本生成模型, 基于 PaLM 2, 8K 上下文, 适合指令跟随与生成任务。

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
Googletext-bison2023-05-108KtexttextProprietary

基准测试表现

基准得分单位备注
MMLU (Massive Multitask Language Understanding)53.9%5-shot
HumanEval45.1pass@1
GSM8K (Grade School Math 8K)42.8%0-shot CoT
MATH12.1%0-shot CoT
BBH (BIG-Bench Hard)48.2%3-shot CoT
GPQA27.7%0-shot
IFEval59.5%prompt_strict
ARC86.7%challenge
MUSR40.8%0-shot
WinoGrande74.1%0-shot

定价

输入输出缓存读取缓存写入

每百万 token

优势

  • 可靠的通用模型。

劣势

  • MMLU 仅 53.9,知识推理偏弱。
  • HumanEval 45.1,代码能力较弱。
  • 闭源专有模型,不支持自托管。
  • 上下文窗口 8K 偏小。

适用场景

  • 通用对话与问答

参考文献