對比 · 2026-06-29
GPT-5 vs Claude Sonnet 4.5 (latest)
並排對比 GPT-5（OpenAI）與 Claude Sonnet 4.5 (latest)（Anthropic）的定價、上下文視窗與能力。 側重：flagship。
| GPT-5 | Claude Sonnet 4.5 (latest) | |
|---|---|---|
| 廠商 | OpenAI | Anthropic |
| 輸入價格 / 1M token | $1.25 | $3.00 |
| 輸出價格 / 1M token | $10.00 | $15.00 |
| 1M token 合計（輸入 + 輸出） | $11.25 | $18.00 |
| 上下文長度 | 400K | 200K |
| 最大輸出 token | 128K | 64K |
| 工具呼叫 | ✓ 是 | ✓ 是 |
| 結構化輸出 | ✓ 是 | 未公開 |
| 推理 | ✓ 是 | ✓ 是 |
| 圖像輸入 | ✓ 是 | ✓ 是 |
| 開放權重 | ✗ 否 | ✗ 否 |
| 渠道可用性 | 25 家渠道 | 21 家渠道 |
| 發布日期 | 2025-08-07 | 2025-09-29 |
| 知識截止 | 2024-09-30 | 2025-07-31 |
一句話結論
- GPT-5 在每百萬 token（輸入 + 輸出合計）成本上便宜約 1.6 倍。
- GPT-5 的上下文視窗更長（400K，對比 200K）。
- 僅 GPT-5 支援結構化輸出 / JSON 模式。
- GPT-5 可在更多提供者處使用（25 vs 21）。
How to read this comparison
This page compares GPT-5 and Claude Sonnet 4.5 (latest) on the dimensions that matter most for production LLM selection: per-token cost, context window, declared capabilities (tool calling, structured output, reasoning, vision), and provider availability.
Green highlights in the table indicate which model leads on a given row. "Leads" means lower price or higher context/capability — not necessarily "better for your use case". A model that costs 3× more may still be the right choice if it unlocks a capability you need.
What this comparison does NOT tell you
- Quality / accuracy — we have no benchmark data. Declared capabilities ≠ measured performance.
- Latency — time-to-first-token varies by provider, region and load. Test with your actual traffic.
- Prompt caching savings — if you reuse system prompts, the cheaper model on headline rate may not be cheapest in practice.
- Fine-tuning availability — not all models can be fine-tuned, even if they are open-weight.
Data is refreshed daily. If a model's capabilities or pricing change, this page updates automatically on the next build cycle.
Frequently asked questions
GPT-5 是否比 Claude Sonnet 4.5 (latest) 便宜？
GPT-5 的輸入價為 $1.25/百萬 token + 輸出價為 $10.00/百萬 token；Claude Sonnet 4.5 (latest) 為 $3.00/百萬 token（輸入）+ $15.00/百萬 token（輸出）。綜合每百萬 token 成本較低的是 GPT-5。
GPT-5 與 Claude Sonnet 4.5 (latest) 誰的上下文更長？
GPT-5 的上下文視窗更長——約 400,000 token vs 200,000 token。
在哪裡可以執行 GPT-5 和 Claude Sonnet 4.5 (latest)？
GPT-5 在 25 家提供者可用；Claude Sonnet 4.5 (latest) 在 21 家提供者可用。完整列表請參見各模型詳情頁。
Explore more
最近更新：
Prices in USD per 1M tokens. Unknown means the provider does not publish per-token pricing.
Pricing and capabilities are refreshed daily and reconciled against each provider's official documentation. Always verify critical production decisions with the provider directly.