Claude Sonnet 4 upgrade enables full codebase processing in a single request

Anthropic has expanded the capabilities of its Claude Sonnet 4 AI model to handle up to one million tokens of context, five times its previous limit, enabling developers to process entire codebases or large document collections in a single request.

The upgrade, now in public beta via Anthropic’s API and Amazon Bedrock, aims to support more complex use cases such as large-scale code analysis, large-scale document synthesis, and context-aware AI agents, with Google Cloud’s Vertex AI integration expected soon.

Anthropic’s move comes as rivals OpenAI and Google market their own AI models with similar context limits, underscoring the race among major providers to handle larger and more complex workloads.

The longer context window removes a key bottleneck for AI-assisted programming by allowing entire projects to be processed at once. Developers no longer need to split large codebases into smaller pieces, reducing the risk of missing links between related components.

Development workflows and staffing

Processing entire codebases in a single AI request could reshape enterprise software development, altering workflows and potentially affecting team structures.

Analysts point out that two trends are driving this shift: model developers are expanding context windows to handle more tokens at once, and AI systems are becoming more capable of accurately processing and reasoning over large volumes of code.

“With expanded context windows, enterprises can potentially accelerate their development and debugging at scale,” said Neil Shah, vice president for research and partner at Counterpoint Research. “Over time, as models become more proficient in generating, validating, and refining boilerplate code, enterprise-grade quality output would be the north star. This gives the enterprise time-to-optimize and time-to-market advantage.”

These performance gains could also change the very nature of a developer’s role, according to Oishi Mazumder, senior analyst at Everest Group.

“Long-context AI moves development from piecemeal assistance to holistic collaboration, turning developers into code orchestrators who direct end-to-end changes across entire systems,” Mazumder said. “This restructuring enables smaller, specialized teams to deliver enterprise-scale projects faster, with gains in onboarding speed, code quality, and delivery pace. The biggest staffing shift will be toward AI-augmented engineers and governance roles, as repetitive coding tasks increasingly move to the AI.”

Security, compliance, and IP risks

AI systems gaining the ability to retain and analyze vast amounts of sensitive code or documents in a single operation could introduce new security, compliance, and safety risks.

“The ability to process entire codebases in one request sharply increases the scale of potential exposure,” Mazumder said. “A single breach could reveal complete system architectures, embedded credentials, and security vulnerabilities at once. Large-context retention also raises compliance risks, as regulated and unregulated data may be mixed, and safety risks, as the AI’s full-system view could be exploited to identify or generate malicious code changes.”

Handling large context inputs adds further complexity, as models process and learn from vast numbers of tokens, said Shah.

“This raises concerns regarding intellectual property (IP) in the generated code, similar to ongoing discussions about AI’s impact on the music industry, where originality and IP rights are sometimes uncertain,” Shah added.