Cursor
@cursor_ai
Claude Opus 4.8 is now available in Cursor. On CursorBench, it's able to work much more efficiently than Opus 4.7. We've also found it to be more persistent on harder tasks.
100retweets2.6klikes
View on XCursor Integrates Claude Opus 4.8 to Boost Agentic Coding Persistence
Cursor, an AI-native code editor, integrated Claude Opus 4.8 as its flagship model. This brings Anthropic's strongest reasoning into the editor's agentic workflow, supporting a 200k token context window that expands to 1M tokens. It features a thinking variant (internal reasoning tokens used to plan before generating code) for complex architectures.
- Context window
- 200K to 1M tokens
- Pricing (input)
- $5 per million tokens
- Pricing (output)
- $25 per million tokens
- Fast mode pricing
- $10 input / $50 output
- Availability
- Live in Cursor editor
The release follows a rapid cycle where Cursor's GPT-5.5 integration recently led internal evaluations. By adding Opus 4.8, Cursor responds to the Claude models' Arena coding leaderboard sweep. The model shows a meaningful gain on CursorBench, specifically in its ability to self-correct during long, autonomous agent sessions.
You can access Opus 4.8 now through the Cursor model selector. While it remains the most expensive tier, a new fast mode is available at $10 per million input tokens—roughly 3x cheaper than Windsurf's Claude Opus 4.7 fast mode integration. Usage draws from your plan's standard API pool.
Still wondering? A few quick answers below.
Claude Opus 4.8 is the latest frontier model from Anthropic integrated into the Cursor code editor. It is designed for autonomous, multi-step agentic coding tasks where the model must hold intent over long sessions. It features a 200k token context window that can expand up to 1M tokens for large codebases.
The thinking variant uses internal reasoning tokens to perform deeper deliberation before producing a final response. This process, known as extended thinking, allows the model to map out complex plans, catch edge cases early, and self-correct when it hits friction. Cursor recommends using the high thinking configuration for the most reliable results on difficult tasks.
Usage is billed through Cursor's API pool at a rate of 5 dollars per million input tokens and 25 dollars per million output tokens. A specialized fast mode is also available for 10 dollars per million input tokens, which offers roughly 2.5 times higher output speeds while remaining significantly cheaper than previous versions.
According to internal CursorBench evaluations, Claude Opus 4.8 shows a meaningful performance jump over the previous 4.7 version. It is more efficient at tool calling and demonstrates higher persistence when tackling messy or ambiguous coding problems. The model is specifically optimized to drive multi-step tasks to completion without losing track of the original goal.
The model has full access to Cursor's agent tools, including semantic search for indexed codebases, file editing, and terminal command execution. It can also control a browser to verify visual changes or take screenshots and perform web searches to fetch real-time information. These capabilities allow it to operate as a self-directed autonomous agent.
