Claude 3 Opus: Complete Benchmark Performance Guide

anthropic 3-opus 2024-03-04 Proprietary

In-depth analysis of Claude 3 Opus performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendoranthropic
Version3-opus
Release Date2024-03-04
Context Window200000 tokens
Input Modalitiestext, image
Output Modalitiestext
LicenseProprietary
Documentationhttps://docs.anthropic.com/claude/docs

Benchmark Performance

Benchmark Score Unit Evaluated At Notes Source
MMLU 81.2 % 2024-03-04 5-shot view
HUMANEVAL 83.3 pass@1 2024-03-04 — view
GSM8K 91.8 % 2024-03-04 0-shot CoT view
MATH 42.7 % 2024-03-04 0-shot CoT view
BBH 81.6 % 2024-03-04 3-shot CoT view
GPQA 46 % 2024-03-04 0-shot view
IFEVAL 79.2 % 2024-03-04 prompt_strict view
ARC 94.5 % 2024-03-04 challenge view
MUSR 53.3 % 2024-03-04 0-shot view
WINOGRANDE 81.9 % 2024-03-04 0-shot view

Pricing

Tier Price Currency
Input$15 / MtokUSD
Output$75 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://www.anthropic.com/pricing · as of 2024-03-04

Compliance

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

Claude 3 Opus

Model Overview

Anthropic Claude 3 Opus 旗舰模型, 200K 上下文, 超越 GPT-4 的推理与创意能力, 适合复杂分析任务。

Core Specifications

Vendor Version Release Date Context Window Input Modalities Output Modalities License
Anthropic 3-opus 2024-03-04 200K text, image text Proprietary

Benchmark Performance

Benchmark Score Unit Notes
MMLU (Massive Multitask Language Understanding) 81.2 % 5-shot
HumanEval 83.3 pass@1 —
GSM8K (Grade School Math 8K) 91.8 % 0-shot CoT
MATH 42.7 % 0-shot CoT
BBH (BIG-Bench Hard) 81.6 % 3-shot CoT
GPQA 46.0 % 0-shot
IFEval 79.2 % prompt_strict
ARC 94.5 % challenge
MUSR 53.3 % 0-shot
WinoGrande 81.9 % 0-shot

Pricing

Input Output Cache Read Cache Write
— — — —

per million tokens

Strengths

  • MMLU score 81.2, strong knowledge reasoning.
  • HumanEval 83.3, excellent code generation.
  • GSM8K 91.8, robust math reasoning.
  • Input Modalities: text, image, audio.
  • Context window 200K.

Weaknesses

  • Proprietary, not self-hostable.

Use Cases

  • Code generation and debugging
  • Long document summarization
  • Vision and image understanding
  • Agent workflows and tool use

References