Archived snippets
Cost
Claude 3 Sonnet has a significant cost advantage over GPT-4-32k. The table below provides a basic comparison, with % differences in brackets:
| Model | Input (USD per 1K tokens) | Output (USD per 1K tokens) |
|---|---|---|
| GPT-4-32k | $0.06 | $0.12 |
| Claude 3 Sonnet | $0.003 (↓ 95%) | $0.015 (↓ 87.5%) |
| Claude 3 Opus | $0.015 (↓ 75%) | $0.075 (↓ 37.5%) |
While Claude 3 is marketed as Anthropic’s middle-tier LLM suitable for RAG applications, Claude 3 Opus is marketed as a top-tier LLM suitable for complex reasoning tasks. Even comparing Claude 3 Opus to GPT-4-32k, we see that GPT-4-32k is significantly more expensive. In our experiments, each question averaged ~6000 input tokens and ~500 output tokens, implying a GPT-4-32k cost of $0.42 against Claude 3 Sonnet cost of $0.025, a 94% reduction in costs.