Codex rate card

Codex rate card

Summary (EN)

OpenAI has published a new Codex rate card that shifts pricing from rough per-message estimates toward a token-based credit system, giving customers a more explicit mapping between usage patterns and cost. The article says most new and existing customers across Plus, Pro, Business, Enterprise, Edu, Health, and Gov plans are being moved to the updated model, while a smaller group of Enterprise accounts remains temporarily on a legacy structure. Under the new system, Codex credits are priced separately for input tokens, cached input tokens, and output tokens, with higher-credit output pricing reflecting the heavier cost of generation. The table covers models including GPT-5.5, GPT-5.4, GPT-5.4 Mini, GPT-5.3-Codex, GPT-5.2, and GPT-Image-2.0, and notes that fast mode can consume credits at higher rates. OpenAI also states that code review uses GPT-5.3-Codex and that GPT-5.3-Codex-Spark may be available as a research preview with non-final rates. Beyond the raw table, the article frames the pricing change as a transparency and workload-shaping move. Instead of treating coding usage as a mostly opaque per-message bundle, OpenAI is aligning Codex with token-metered economics similar to API thinking, which should make it easier for organizations to model cost drivers such as large contexts, cached prompts, output-heavy tasks, and high-speed execution modes. For application builders and enterprise buyers, this is operationally important: it suggests the economics of coding agents are maturing into a more infrastructure-like pricing model, where orchestration patterns and workload design directly determine spend.

Summary (ZH)

OpenAI 发布了新的 Codex 价格卡，把原先较为粗略的“按消息估算成本”方式，转向更明确的基于 token 的积分计费体系，使客户能够更直接地把使用模式映射到实际成本。文章称，大多数 Plus、Pro、Business、Enterprise、Edu、Health 和 Gov 用户都将迁移到这一新模式，只有少量 Enterprise 账户会暂时继续使用旧版计费结构。在新体系下，Codex 按输入 token、缓存输入 token 和输出 token 分别计费，其中输出部分的积分价格更高，反映出生成成本更重。价格表覆盖 GPT-5.5、GPT-5.4、GPT-5.4 Mini、GPT-5.3-Codex、GPT-5.2 以及 GPT-Image-2.0 等模型，并说明 fast mode 在支持的模型上会以更高倍率消耗积分。OpenAI 还明确指出，代码审查场景使用 GPT-5.3-Codex，而 GPT-5.3-Codex-Spark 可能以研究预览形式提供，费率尚未最终确定。除了价格表本身，这次调整更像一次透明化和工作负载重塑。OpenAI 不再把编码代理的成本包装为模糊的“每条消息大概多少钱”，而是把 Codex 拉向类似 API 的 token 计量逻辑，使企业更容易识别上下文长度、缓存命中、输出规模和高速执行模式等成本驱动因素。对应用开发者和企业采购方来说，这具有很强的操作意义，因为它表明编码代理的商业模式正在从“聊天工具收费”转向更接近基础设施的计费框架，编排方式和任务设计本身将更直接地决定支出水平。

Source

https://help.openai.com/en/articles/20001106-codex-rate-card