OpenAI 发布 GPT-6.1 Sol 模型，性能媲美旗舰款价格仅为五分之一

OpenAI 在 9 月 30 日的开发者活动日中正式发布 GPT-6.1 Sol 模型 1。该模型在性能接近旗舰级 GPT-6 Astra 的同时，标准输入输出 token 价格仅为后者的五分之一 1。根据官方数据，GPT-6.1 Sol 在软件工程任务中与 GPT-6 Astra 表现相当，在电脑操作任务中得分仅低 Astra 2.1 个百分点，但单任务成本约为 Astra 的七分之一 1。

新模型在可靠性方面也有明显改进 1。事实错误率从 GPT-6 Sol 的 11.4% 降至 7.7%，下降幅度约为 32% 1。缓存费用方面，每 100 万 Token 缓存费用为 0.1 美元，相比 GPT-6 Sol 低 50% 1。该模型在智能体编程、电脑操作、专业工作流等任务上表现出众 1。

GPT-6.1 Sol 已从 9 月 29 日起向 ChatGPT Plus、Pro、Business、Enterprise 和 Edu 用户开放 1。开发者可通过 API 以 gpt-6.1-sol 调用该模型 1。

OpenAI unveiled GPT-6.1 Sol on September 30 during its developer event, positioning the new model as the most cost-effective option in its current lineup 1. The model achieves performance comparable to the flagship GPT-6 Astra while charging only one-fifth of Astra's standard input and output token prices 1.

GPT-6.1 Sol demonstrates strong capabilities across multiple domains, performing equivalently to Astra on software engineering tasks and falling just 2.1 percentage points behind on computer operation benchmarks—while costing approximately one-seventh as much per task 1. The model also features improved reliability, reducing hallucination rates from GPT-6 Sol's 11.4 percent to 7.7 percent, a reduction of roughly 32 percent 1. Caching costs have been cut by 50 percent compared to GPT-6 Sol, now standing at $0.10 per million tokens 1.

The new model is available to ChatGPT Plus, Pro, Business, Enterprise, and Edu subscribers, with developers able to access it through the API using the identifier gpt-6.1-sol 1.