November 2025 Summaries

Anthropic's latest model, Claude Opus 4.5, is now available for Warp users, optimized for long-horizon tasks and designed to work seamlessly with Warp's new planning entry point, /plan. This integration allows for creating persistent plans that serve as structured documentation of the model's reasoning and decisions, enhancing its performance in autonomous, multi-step workflows. The harness for Claude Opus 4.5 has been refined to differentiate clearly between planning and execution phases, leveraging the model's improved memory and state management to make transitions from high-level plans to concrete todo lists more intuitive. These enhancements have resulted in a 15% performance improvement on the T-Bench compared to the previous Claude 4.5 Sonnet model. The new planning mode allows for dynamic plan creation and updates, facilitating live collaboration between the agent and the user, with plans being saved as durable documents on Warp Drive for future reference and use in collaborative environments such as pull requests.

Warp introduces new features that enhance the capabilities of its agents, allowing them to interact with terminal applications in real-time, plan and review code collaboratively, and integrate seamlessly with tools like Slack, Linear, and GitHub Actions. With Full Terminal Use, agents can engage with interactive CLI apps, such as debuggers and database shells, enabling real-time task execution and problem-solving without background processes. The /plan feature facilitates spec-driven development by aligning the user and agent on a shared implementation strategy, allowing for versioned and shareable planning documents. Interactive Code Review empowers developers to oversee and refine the agent's output directly within the terminal, ensuring human oversight and control over code quality. Warp's first-party integrations allow agents to initiate tasks from within popular tools, providing real-time visibility and collaborative session sharing, all while maintaining security through sandboxed environments and controlled permissions. These innovations aim to make Warp's agent more effective, reliable, and collaborative, marking the beginning of further platform development.

Google's latest model, Gemini 3 Pro, has been made available to all Warp users and has already impressed the Warp team with its performance. The model demonstrated over a 15% improvement on the Terminal-Bench 2.0, raising Warp's score from 50.1% to 59.1%, positioning it second on the board. The Warp team, having tested Gemini 3 Pro internally for about a week, has observed standout performance in various tests and shared positive feedback via their Slack channel.

Support for OpenAI's latest model, GPT-5.1, is now available in Warp, becoming the default model for new users due to its impressive capabilities demonstrated during extensive testing. GPT-5.1 is an enhancement of GPT-5, offering significant improvements in speed through adaptive reasoning, which allows the model to adjust its reasoning based on task complexity. This results in a 15% faster time to first token and 40% faster task completion, while maintaining high-quality output and excelling in deep debugging scenarios. The development of GPT-5.1 in collaboration with OpenAI included optimizing prompt structure and tool definitions, ensuring a balance between planning and execution within the terminal. To encourage developers to try GPT-5.1, Warp is offering a promotional code for a discounted first month of Warp Build to the first 510 users.