IT之家 9 月 30 日消息,在今天召开的 OpenAI 开发者活动日中,官方正式宣布推出 GPT-6.1 Sol 模型,在表现接近 Astra 的同时,其费用仅为 Astra 的五分之一,官方称这是“在当下同等表现下性价比最高的模型”。
GPT‑6 Astra 是 OpenAI 在 2026 年 9 月推出的旗舰级大语言模型,官方定位是“迄今最智能、最对齐(最符合人类意图)”的模型,重点面向长程多步骤任务与专业工作流。
费用维度,依据官网公布的费用细节,IT之家附上相关售价信息如下:
gpt-6.1-sol 全模式定价表(短上下文,输入 tokens ≤ 272,000)
| 模式 | 输入 (Input) | 缓存输入 (Cached input) | 缓存写入 (Cache writes) | 输出 (Output) |
|---|---|---|---|---|
| Standard(标准) | $2.00 | $0.10 | $2.50 | $10.00 |
| Batch(批量) | $1.00 | $0.05 | $1.25 | $5.00 |
| Flex(弹性) | $1.00 | $0.05 | $1.25 | $5.00 |
| Fast(快速) | $4.00 | $0.20 | $5.00 | $20.00 |
gpt-6.1-sol 全模式定价表(长上下文,输入 tokens > 272,000)
| 模式 | 输入 (Input) | 缓存输入 (Cached input) | 缓存写入 (Cache writes) | 输出 (Output) |
|---|---|---|---|---|
| Standard(标准) | $4.00 | $0.20 | $5.00 | $15.00 |
| Batch(批量) | $2.00 | $0.10 | $2.50 | $7.50 |
| Flex(弹性) | $2.00 | $0.10 | $2.50 | $7.50 |
| Fast(快速) | $8.00 | $0.40 | $10.00 | $30.00 |
上述 4 种处理模式区别如下:
| 模式 | 处理速度 | 适用情形 |
|---|---|---|
| Standard(标准) | 正常速度 | 日常开发、一般应用 |
| Batch(批量) | 较慢(异步批量处理) | 非实时、大批量任务(如离线数据处理) |
| Flex(弹性) | 可变(依据负载动态调整) | 需要成本与速度平衡的情形 |
| Fast(快速) | 最快(优先处理) | 对延迟敏感的应用(如实时对话、低延迟响应) |
性价比维度,官方表示 GPT-6.1 Sol 模型每 100 万 Token 的缓存费用为 0.1 美元(IT之家注:现汇率约合 0.67 元人民币),相比较 GPT-6 Sol 模型定价低 50%。
在表现维度,在智能体编程、电脑操作和专业工作流等任务上,表现接近旗舰 GPT‑6 Astra,但标准输入 / 输出 token 售价只有 Astra 的五分之一。
除了“能干活”,GPT‑6.1 Sol 在可靠性与对齐维度也有改进:
调用维度,自 9 月 29 日起向 ChatGPT Work 与 Codex 的 Plus / Pro / Business / Enterprise / Edu 使用者开放;开发者可借助 API 以 gpt-6.1-sol 调用。
本文来源:https://www.ithome.com/1/008/527.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
IT Home reported on Sep 30 that, at the OpenAI developer event held today, the company officially announced the GPT-6.1 Sol model, which delivers performance close to Astra while costing only one-fifth as much as Astra. The official stated that this is “the most cost-effective model today at comparable performance.”
GPT-6 Astra is OpenAI’s flagship large language model launched in September 2026, officially positioned as “the most intelligent and most aligned (most consistent with human intent) model to date,” with a focus on long-horizon multi-step tasks and professional workflows.
On the pricing dimension, based on the pricing details published on the official website, IT Home provides the relevant pricing information below:
gpt-6.1-sol full-mode pricing table (short context, input tokens ≤ 272,000)
gpt-6.1-sol full-mode pricing table (long context, input tokens > 272,000)
The differences among the four processing modes above are as follows:
On the cost-effectiveness dimension, the official stated that the GPT-6.1 Sol model’s cache cost is $0.1 per 1,000,000 tokens (IT Home note: approximately RMB 0.67 at the current exchange rate), which is 50% lower than the pricing of the GPT-6 Sol model.
On the performance dimension, for tasks such as agentic programming, computer operation, and professional workflows, its performance approaches that of the flagship GPT-6 Astra, while the standard input / output token price is only one-fifth that of Astra.
Beyond “being able to do work,” GPT-6.1 Sol also has improvements in the reliability and alignment dimensions:
On the access dimension, starting Sep 29, it is open to Plus / Pro / Business / Enterprise / Edu users of ChatGPT Work and Codex; developers can call it via the API as gpt-6.1-sol.



