Anthropic announced Claude Opus 4.8, an updated flagship model focused on code reliability, long-session memory, and agent stability.

Key improvements

The new release aims to improve developer-facing behaviour, offering more consistent results in multi-step tasks and reduced self-generated errors.

• SWE-Bench Pro score increased to 69.2% compared with 64.3% for the previous version.
• Better context retention across extended sessions, which supports longer uninterrupted dialogues and workflows.
• Fewer mistakes in generated code and in the model’s internal evaluation of its outputs.
• Claude Code now includes a Dynamic Workflows mode that launches multiple parallel subagents for complex tasks.
• Fast Mode is reported to be approximately 2.5× faster and 3× cheaper compared with its prior configuration.

Benchmarks and behaviour

According to Anthropic, the jump on SWE-Bench Pro reflects concrete improvements in software engineering tasks, particularly for multi-step reasoning and code synthesis.

The company highlights gains in memory management and agent coordination, which should reduce the need for manual intervention in extended sessions.

Availability and pricing

Anthropic states that the usage price of Claude Opus 4.8 remains unchanged, while performance and efficiency enhancements are intended to lower effective operating costs for users.

Organizations and developers can evaluate the model through Anthropic’s existing access channels; the company positions this release as an incremental upgrade focused on reliability.