GPT-6.1 Sol 登陆 Amazon Bedrock，为日常工作带来接近 Astra 的智能水平
英文标题：Bring near-Astra intelligence to everyday work with GPT-6.1 Sol on Amazon Bedrock
GPT-6.1 Sol 现已在 Amazon Bedrock 正式可用，面向高频运行的编程、计算机操作和专业工作负载提供更强推理能力。OpenAI 称，该模型在 DeepSWE v1.1 上以约五分之一的单任务成本达到 GPT-6 Astra 的水平，并较 GPT-6 Sol 的最佳成绩高出 6.4 个百分点。开发者可通过 Amazon Bedrock API、Codex 及 AWS Agent Toolkit 将其用于代码开发、文档分析和跨业务工具的多步骤工作流。Amazon Bedrock 同时提供 IAM、CloudTrail、VPC 端点及零运营人员访问等生产环境治理与数据保护能力。
1.3 MB
101.1 KB
124.1 KB
345.0 KB
141.2 KB
346.5 KB
65.4 KB
73.2 KB
22.6 KB
229.6 KB
中文
GPT-6.1 Sol 现已在 Amazon Bedrock 正式可用，为高频运行的编程、计算机操作和专业工作负载带来更强的推理能力。
要让 AI Agent 完成一项任务，它可能需要收集信息、使用工具、测试不同方法、从错误中恢复，并验证结果。每一个决策都会影响后续进展。一次错误的转向，可能会在 Agent 获得有用结果前增加模型交互次数、工具调用次数、延迟以及人工干预。
AI Agent 的经济性贯穿整个任务流程。Token 价格影响每次交互的成本，而推理质量则影响完成工作需要多少次交互，以及这些交互能否导向成功结果。完成一项任务的总成本取决于两者。
GPT-6.1 Sol 已在 Amazon Bedrock 正式可用，运行于专为大规模性能、安全性和可靠性打造的推理引擎之上。作为 GPT-6 Sol 的重大升级，它在 Agentic 编程、计算机操作和专业工作方面表现强劲。根据 OpenAI 的说法，它为日常工作流带来了接近 Astra 的智能水平。
将更强推理能力应用于 Agentic 工作
更高效地驾驭软件工程工作流
软件工程展示了推理质量如何影响整个工作流。一个 Agent 可能需要理解陌生代码库、追踪依赖关系、判断应在何处进行修改，并验证实现结果。根据 OpenAI 的数据，GPT-6.1 Sol 在 DeepSWE v1.1 上以约五分之一的单任务成本达到 GPT-6 Astra 的水平。它还在比该 GPT-6 Sol 结果更低的推理强度下，超过 GPT-6 Sol 的最佳得分 6.4 个百分点。
Codex 可将这一推理能力应用于完整的软件开发周期。你可以将 Codex 配置为使用 Amazon Bedrock 上的 GPT-6.1 Sol，执行涵盖调研、实现和测试的工作。Codex 可与代码库、本地文件、终端和开发工具协同，编写功能、修复 Bug 并运行测试。你可以通过桌面应用、CLI 和受支持的 IDE 使用 Codex。对于 AWS 开发，AWS Agent Toolkit 可通过一条终端命令，将 Codex 连接至 AWS 文档、API 和服务。
当 Agent 必须解读复杂文档、选择适当工具并随条件变化而调整时，同样可以应用这些推理能力。根据 OpenAI 的说法，GPT-6.1 Sol 在复杂文档分析方面接近 GPT-6 Astra，并且在跨业务工具完成多步骤工作流时优于 GPT-6 Sol。
你可以通过开箱即用的体验或自行构建的应用来应用这些能力。在桌面应用中，ChatGPT Work 可以跨文件和应用收集信息，并将其转化为完成的交付成果。借助受支持的 Amazon Bedrock API，你还可以构建用于综合文档的内部工具、协调跨系统工作的 Agent，以及评估多项输入的面向客户应用。
识别局限并遵守约束
要在多种工具间高效工作，Agent 还需要识别工具何时发生故障、某项操作受到限制，或信息缺失。沟通这些局限性，可以让应用或用户在 Agent 基于不完整信息继续执行，或采取非预期操作之前进行干预。根据 OpenAI 的说法，GPT-6.1 Sol 在透明度、用户意图和明确限制等高难度评估中优于 GPT-6 Sol。
使用受支持的 Amazon Bedrock API 构建应用时，你可以定义模型可用的工具。你还可以决定当某项操作需要批准或无法完成时，应用应如何响应。这些应用层控制与模型改进相辅相成，可帮助你在工作流到达具有重要影响的决策点时，持续让人员参与其中。
在 Amazon Bedrock 上运行 GPT-6.1 Sol
Amazon Bedrock 提供了在生产环境运行 GPT-6.1 Sol 所需的基础设施和控制能力。你可以通过 AWS Identity and Access Management（IAM）策略治理模型访问权限，并通过 AWS CloudTrail 审计调用。为帮助将流量保持在你的网络边界内，你可以使用由 AWS PrivateLink 支持的虚拟私有云（VPC）终端节点。推理在硬件隔离的基础设施上运行，且运营人员零访问权限，因此即使是 AWS 运营人员也无法在推理期间访问你的提示词和生成结果。
你的推理数据不会用于模型训练，使用 GPT-6.1 Sol 也不要求你选择与 OpenAI 共享数据。对于自动化滥用检测，被分类器标记的流量会由 AWS 最多保留 30 天，并以程序化方式处理。你可以通过 AWS 客户团队申请零数据保留。详情请参阅数据保留。
开始使用
GPT-6.1 Sol 以更低的单任务成本为 Agentic 工作带来更强推理能力，使你的 Agent 能以更少步骤得到正确答案。你可以在 Amazon Bedrock 控制台中开始使用 GPT-6.1 Sol，或通过受支持的 Amazon Bedrock API 以编程方式使用。有关受支持的 AWS 区域、终端节点、API、功能、推理配置文件和定价的信息，请参阅 Amazon Bedrock 文档。
想了解 Amazon Bedrock 如何支持你的团队？ 联系我们，开启沟通。
关于作者
Tanvi Girinath
Tanvi 是 Amazon Web Services（AWS）Amazon Bedrock 的产品营销经理，帮助客户借助 Amazon Bedrock 采用并规模化部署 AI 应用和 Agent。
Chris Dickens
Chris 是 OpenAI 产品团队成员，专注于 OpenAI API。他的工作包括与 AWS 合作，通过 Amazon Bedrock 让开发者能够广泛使用 OpenAI 的前沿模型。
Manish Rathaur
Manish 是 Amazon Bedrock 高级产品经理。
English / 原文
GPT-6.1 Sol is now generally available on Amazon Bedrock, bringing stronger reasoning to coding, computer use, and professional workloads that run frequently.
For an AI agent to complete a task, it may need to gather information, use tools, test different approaches, recover from errors, and verify its result. Every decision shapes what happens next. A wrong turn can add model interactions, tool calls, latency, and human intervention before the agent reaches a useful result.
The economics of an AI agent take shape across the entire task. Token prices influence the cost of each interaction, while reasoning quality influences how many interactions the work requires and whether they lead to a successful outcome. The total cost of completing a task depends on both.
generally available on Amazon Bedrock, running on an inference engine built for performance, security, and reliability at scale. A major upgrade to GPT-6 Sol, it delivers strong performance on agentic coding, computer use, and professional work. According to OpenAI, it brings near-Astra intelligence to everyday workflows.
Apply stronger reasoning across agentic work
Navigate software engineering workflows more effectively
Software engineering shows how reasoning quality affects an entire workflow. An agent may need to understand an unfamiliar repository, trace dependencies, determine where to make a change, and validate the implementation. According to OpenAI, GPT-6.1 Sol matches GPT-6 Astra on DeepSWE v1.1 at roughly one-fifth the cost per task. It also exceeds the best score from GPT-6 Sol by 6.4 percentage points while using a lower reasoning effort than that GPT-6 Sol result.
Codex puts that reasoning to work across the full development cycle. You can configure Codex to use GPT-6.1 Sol on Amazon Bedrock for work spanning investigation, implementation, and testing. Codex works with repositories, local files, terminals, and development tools to write features, fix bugs, and run tests. You can access Codex through the desktop app, CLI, and supported IDEs. For AWS development, the Agent Toolkit for AWS connects Codex to AWS documentation, APIs, and services through a single terminal command.
The same reasoning capabilities apply when agents must interpret complex documents, select the appropriate tools, and adapt as conditions change. According to OpenAI, GPT-6.1 Sol approaches GPT-6 Astra on complex document analysis and improves on GPT-6 Sol when completing multistep workflows across business tools.
You can apply these capabilities through ready-to-use experiences or applications you build. In the desktop app, ChatGPT Work can gather information across files and applications and turn it into finished deliverables. Using supported Amazon Bedrock APIs, you can also build internal tools that synthesize documents, agents that coordinate work across systems, and customer-facing applications that evaluate multiple inputs.
Recognize limitations and respect constraints
Working effectively across tools also requires an agent to recognize when a tool has failed, an action is restricted, or information is missing. Communicating these limitations allows the application or user to intervene before the agent continues with incomplete information or takes an unintended action. According to OpenAI, GPT-6.1 Sol improves on GPT-6 Sol in challenging evaluations of transparency, user intent, and explicit restrictions.
When building with supported Amazon Bedrock APIs, you can define the tools available to the model. You can also determine how your application responds when an action requires approval or cannot be completed. These application-level controls complement the model improvements by helping you keep people involved when a workflow reaches a consequential decision.
Run GPT-6.1 Sol on Amazon Bedrock
Amazon Bedrock provides the infrastructure and controls to run GPT-6.1 Sol in production. You can govern model access through AWS Identity and Access Management (IAM) policies and audit invocations through AWS CloudTrail. To help keep traffic within your network boundaries, you can use virtual private cloud (VPC) endpoints powered by AWS PrivateLink. Inference runs on hardware-isolated infrastructure with zero-operator access, so not even AWS operators can access your prompts and completions during inference.
Your inference data isn’t used for model training, and using GPT-6.1 Sol doesn’t require you to opt into sharing your data with OpenAI. For automated abuse detection, classifier-flagged traffic is retained by AWS for up to 30 days and processed programmatically. You can request zero data retention through your AWS account team. See data retention for details.
Get started
GPT-6.1 Sol brings stronger reasoning to agentic work at a fraction of the cost per task, so your agents reach the right answer in fewer steps. You can get started with GPT-6.1 Sol in the Amazon Bedrock console or programmatically through supported Amazon Bedrock APIs. For information about supported AWS Regions, endpoints, APIs, features, inference profiles and pricing, see the Amazon Bedrock documentation.
Interested in how Amazon Bedrock can support your team? Connect with us to start the conversation.
About the authors
Tanvi Girinath
Tanvi is a Product Marketing Manager for Amazon Bedrock at Amazon Web Services (AWS), where she helps customers adopt and scale AI applications and agents with Amazon Bedrock.
Chris Dickens
Chris is a Member of Product Staff at OpenAI focused on the OpenAI APIs. His work includes collaboration with AWS on Amazon Bedrock to make OpenAI’s frontier models widely accessible to developers.
Manish Rathaur
Manish is a Senior Product Manager for Amazon Bedrock.