GPT-6.1 Sol is now generally available on Amazon Bedrock. It is built for performance, security, and reliability at scale. This upgrade delivers better reasoning for tasks like coding, document analysis, and workflows.
In software engineering, GPT-6.1 Sol helps understand repositories, trace dependencies, and validate changes. It matches GPT-6 Astra on DeepSWE v1.1 at about 20% of the cost per task. It also beats GPT-6 Sol by 6.4 percentage points while using less reasoning effort.
Codex can use GPT-6.1 Sol across the full development cycle. It works with repositories, files, terminals, and tools to write code, fix bugs, and run tests. Developers can access Codex through desktop apps, command line, and IDEs. The Agent Toolkit for AWS connects Codex to AWS services with a single command.
The reasoning capabilities also help when agents interpret complex documents, choose tools, and adapt to changing conditions. GPT-6.1 Sol improves on GPT-6 Sol in document analysis and multi-step workflows. It can be used to build internal tools, customer apps, and workflows that synthesize information.
Building effective agents requires recognizing when tools fail or restrictions apply. GPT-6.1 Sol improves transparency and handling of restrictions. You can define tools and control actions through Amazon Bedrock APIs. You can also set policies for model access and audit usage.
Amazon Bedrock provides the infrastructure to run GPT-6.1 Sol securely. It uses hardware-isolated infrastructure with no operator access. Data used during inference is not shared with OpenAI, and you can request no data retention. Traffic can be kept within your network using VPC endpoints.
GPT-6.1 Sol offers stronger reasoning at a lower cost per task. It helps agents reach correct answers faster and more reliably. You can start using GPT-6.1 Sol via the Amazon Bedrock console or APIs. For details on regions, endpoints, and pricing, see the documentation.
