OpenAI
GPT-5 API
OpenAI's GPT-5 line, priced per million tokens, with a near eightfold spread.
4
个模型
The GPT-5 line covers OpenAI's general-purpose models, and the four tiers we serve are separated mainly by how much reasoning you are paying for. All of them speak the OpenAI protocol, so they are drop-in for existing OpenAI client code, and all support prompt caching — cached input is billed at a tenth of fresh input, which is the single biggest lever on cost for repeated system prompts.
概览
- 起价
- $0.195 / 100 万 tokens
该系列的模型
| 模型 | 起价 | 模式 | 打开 |
|---|---|---|---|
| GPT 5.6 LunaFast, cost-efficient model of the GPT-5.6 series. Suited for high-volume, latency-sensitive chat, classification and lightweight agentic workflows. | $0.195/ 100 万 tokens | — | 打开文档 |
| GPT 5.6 TerraHigh-capability GPT-5.6 model for long-context, tool-enabled workflows across writing, coding, analysis and multi-step problem solving. Text, image and file input. | $0.6/ 100 万 tokens | — | 打开文档 |
| GPT 5.5Broadly capable GPT-5.5 model for everyday reasoning, coding and writing. | $1.08/ 100 万 tokens | — | 打开文档 |
| GPT 5.6 SolFrontier model of the GPT-5.6 series for the hardest reasoning, coding and agentic tasks. Text, image and file input, 1M-token context. | $1.5/ 100 万 tokens | — | 打开文档 |
价格
| 模型 | 选项 | 价格 |
|---|---|---|
| GPT 5.6 Luna | input | $0.195/ 100 万 tokens |
| GPT 5.6 Luna | output | $1.17/ 100 万 tokens |
| GPT 5.6 Luna | cacheRead | $0.019/ 100 万 tokens |
| GPT 5.6 Luna | cacheWrite | $0.243/ 100 万 tokens |
| GPT 5.6 Terra | input | $0.6/ 100 万 tokens |
| GPT 5.6 Terra | output | $3.6/ 100 万 tokens |
| GPT 5.6 Terra | cacheRead | $0.06/ 100 万 tokens |
| GPT 5.6 Terra | cacheWrite | $0.75/ 100 万 tokens |
| GPT 5.5 | input | $1.08/ 100 万 tokens |
| GPT 5.5 | output | $6.47/ 100 万 tokens |
| GPT 5.5 | cacheRead | $0.108/ 100 万 tokens |
| GPT 5.6 Sol | input | $1.5/ 100 万 tokens |
| GPT 5.6 Sol | output | $9/ 100 万 tokens |
| GPT 5.6 Sol | cacheRead | $0.15/ 100 万 tokens |
| GPT 5.6 Sol | cacheWrite | $1.88/ 100 万 tokens |
如何选择
GPT 5.6 Luna
High-volume, latency-sensitive work — chat, classification, light agentic steps. Roughly an eighth of Sol's input price.
GPT 5.6 Terra
Long-context, tool-enabled workflows across writing, coding and analysis. The middle of the line.
GPT 5.5
The previous generation, for everyday reasoning, coding and writing.
GPT 5.6 Sol
The hardest reasoning, coding and agentic tasks, with a 1M-token context and file input.
优势
- Cached input is billed at one tenth of fresh input on every tier, which pays back immediately on a fixed system prompt.
- One protocol across the whole line — moving between tiers is a model-name change.
- Sol carries a 1M-token context, so long documents do not need chunking.
- Luna makes high-volume classification and extraction genuinely cheap within the same family.
限制
- Output tokens cost roughly six times input on every tier, so verbose responses dominate the bill.
- Only the 5.6 tiers offer cache writes; on 5.5 you pay full price to populate the cache.
- Text-only in and out here — image and video generation come from other families.
厂商对该系列的官方说明： OpenAI — models