Claude 3.5 Sonnet (2024-10-22): Complete Benchmark Performance Guide

anthropic 3.5-sonnet-20241022 2024-10-22 Proprietary

In-depth analysis of Claude 3.5 Sonnet (2024-10-22) performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendoranthropic
Version3.5-sonnet-20241022
Release Date2024-10-22
Context Window200000 tokens
Input Modalitiestext, image
Output Modalitiestext
LicenseProprietary
Documentationhttps://docs.anthropic.com/claude/docs

Benchmark Performance

BenchmarkScoreUnitEvaluated AtNotesSource
MMLU87.2%2024-10-225-shotview
HUMANEVAL86.1pass@12024-10-22—view
GSM8K94%2024-10-220-shot CoTview
MATH75.1%2024-10-220-shot CoTview
BBH84.2%2024-10-223-shot CoTview
GPQA50.9%2024-10-220-shotview
IFEVAL81.6%2024-10-22prompt_strictview
ARC96.8%2024-10-22challengeview
MUSR71.8%2024-10-220-shotview
WINOGRANDE86.9%2024-10-220-shotview

Pricing

TierPriceCurrency
Input$3 / MtokUSD
Output$15 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://www.anthropic.com/pricing · as of 2024-10-22

Compliance

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

Claude 3.5 Sonnet (2024-10-22)

Modellöversikt

Anthropic Claude 3.5 Sonnet 2024-10-22 版本, 引入计算机使用能力, 编码与推理性能再次提升。

Kärnspecifikationer

LeverantörVersionReleasedatumKontextfönsterInmatningsmodaliteterUtmatningsmodaliteterLicens
Anthropic3.5-sonnet-202410222024-10-22200Ktext, imagetextProprietary

Benchmarkprestanda

BenchmarkPoängEnhetAnteckningar
MMLU (Massive Multitask Language Understanding)87.2%5-shot
HumanEval86.1pass@1—
GSM8K (Grade School Math 8K)94.0%0-shot CoT
MATH75.1%0-shot CoT
BBH (BIG-Bench Hard)84.2%3-shot CoT
GPQA50.9%0-shot
IFEval81.6%prompt_strict
ARC96.8%challenge
MUSR71.8%0-shot
WinoGrande86.9%0-shot

Priser

InmatningUtmatningCacheläsningCacheskrivning
————

per miljon tokens

Styrkor

  • MMLU score 87.2, strong knowledge reasoning.
  • HumanEval 86.1, excellent code generation.
  • GSM8K 94.0, robust math reasoning.
  • 上下文窗口 200K，支持长文本。

Svagheter

  • 闭源专有模型，不支持自托管。

Användningsområden

  • 代码生成与调试
  • 长文档摘要
  • Agent 工作流与工具调用

Referenser