Key Specifications

Vendorgoogle
Version1.0-ultra
Release Date2023-12-06
Context Window32768 tokens
Input Modalitiestext, image
Output Modalitiestext
LicenseProprietary
Documentationhttps://ai.google.dev/gemini-api/docs

Benchmark Performance

BenchmarkScoreUnitEvaluated AtNotesSource
MMLU80.3%2023-12-065-shotview
HUMANEVAL81.7pass@12023-12-06view
GSM8K85.9%2023-12-060-shot CoTview
MATH57.1%2023-12-060-shot CoTview
BBH83.8%2023-12-063-shot CoTview
GPQA46.9%2023-12-060-shotview
IFEVAL81.6%2023-12-06prompt_strictview
ARC95.7%2023-12-06challengeview
MUSR64.3%2023-12-060-shotview
WINOGRANDE80.4%2023-12-060-shotview

Pricing

TierPriceCurrency
Input$7 / MtokUSD
Output$21 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://ai.google.dev/pricing · as of 2023-12-06

Compliance

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

Gemini 1.0 Ultra

Modellübersicht

Google Gemini 1.0 Ultra 旗舰模型, 32K 上下文, 在多项基准上接近 GPT-4, 多模态推理能力突出。

Kernspezifikationen

AnbieterVersionVeröffentlichungsdatumKontextfensterEingabemodalitätenAusgabemodalitätenLizenz
Google1.0-ultra2023-12-0632Ktext, imagetextProprietary

Benchmark-Leistung

BenchmarkErgebnisEinheitNotizen
MMLU (Massive Multitask Language Understanding)80.3%5-shot
HumanEval81.7pass@1
GSM8K (Grade School Math 8K)85.9%0-shot CoT
MATH57.1%0-shot CoT
BBH (BIG-Bench Hard)83.8%3-shot CoT
GPQA46.9%0-shot
IFEval81.6%prompt_strict
ARC95.7%challenge
MUSR64.3%0-shot
WinoGrande80.4%0-shot

Preise

EingabeAusgabeCache-LesenCache-Schreiben

pro Million Token

Stärken

  • MMLU score 80.3, strong knowledge reasoning.
  • HumanEval 81.7, excellent code generation.
  • GSM8K 85.9, robust math reasoning.
  • 支持文本、图像、音频多模态输入。

Schwächen

  • 闭源专有模型,不支持自托管。

Anwendungsfälle

  • 代码生成与调试
  • 视觉与图像理解
  • Agent 工作流与工具调用

Referenzen