Overview
Cursor has significantly increased usage limits for their Auto and Composer 1.5 models across all individual plans by introducing two separate usage pools. The update includes their custom-trained Composer 1.5 model, which scores above Sonnet 4.5 on Terminal-Bench 2.0, and provides 3x the usage limit of Composer 1 (temporarily 6x through February 16).
What You'll Learn
1
How Cursor's new dual usage pool system works for Auto + Composer vs API models
2
Why Cursor trained their own Composer 1.5 model for agentic coding instead of relying solely on frontier models
3
How Composer 1.5 benchmarks against frontier models like Sonnet 4.5 on Terminal-Bench 2.0
4
How to monitor and manage usage across the two different usage pools in Cursor's editor
Prerequisites & Requirements
- Cursor editor with an individual plan (Pro, Pro Plus, or Ultra)
- Basic familiarity with AI-assisted coding tools and agentic coding workflows(optional)
Key Questions Answered
What are the two usage pools in Cursor's updated pricing model?
Cursor now separates usage into two pools: Auto + Composer, which includes significantly more usage for the Auto model selector and Composer 1.5 model, and API, which charges the standard API price of whichever model you select. Individual plans include at least $20 of API usage monthly with the option to pay for additional usage.
How does Composer 1.5 compare to Sonnet 4.5 on Terminal-Bench 2.0?
Composer 1.5 scores above Sonnet 4.5 on Terminal-Bench 2.0, an agent evaluation benchmark for terminal use maintained by the Laude Institute. However, it scores below the best frontier models. Cursor's score was computed using the official Harbor evaluation framework with default benchmark settings, running 2 iterations per model-agent pair and reporting the average.
How much more usage does Composer 1.5 get compared to Composer 1?
Composer 1.5 has 3x the usage limit of Composer 1 as a permanent increase. For a limited promotional period through February 16, this limit is further boosted to 6x more than Composer 1. These increased limits apply to all individual plans including Pro, Pro Plus, and Ultra.
Which Cursor plans benefit from the increased agent usage limits?
The increased usage limits are now live for all individual plans: Pro, Pro Plus, and Ultra. Both usage pools reset with each user's monthly usage cycle. The API usage pool includes at least $20 of usage each month, with higher tiers receiving more, while Auto + Composer usage is significantly expanded.
Why did Cursor train their own model for agentic coding?
Training their own Composer 1.5 model allows Cursor to include significantly more usage in a sustainable way. As developers shifted toward using agents for ambitious, codebase-wide changes, Cursor needed a cost-effective model that balances speed, intelligence, and affordability alongside the latest frontier models available through API pricing.
How can I monitor my Cursor usage across the two pools?
Cursor has added a new usage settings page directly in the editor where you can monitor your limits across the two usage pools. The page shows Auto + Composer usage separately from API usage, giving visibility into how much of each pool you've consumed during your billing cycle.
Key Statistics & Figures
Composer 1.5 usage limit increase over Composer 1
3x
Permanent increase for all individual plans
Temporary promotional usage limit increase
6x
Available through February 16, compared to Composer 1
Minimum monthly API usage included
$20
For individual plans, with more on higher tiers
Terminal-Bench 2.0 evaluation iterations
2 iterations per model-agent pair
Used to compute Composer 1.5's benchmark score, reporting the average
Technologies & Tools
IDE/Editor
Cursor
AI-powered code editor with agentic coding capabilities
AI Model
Composer 1.5
Cursor's custom-trained model for agentic coding, benchmarked on Terminal-Bench 2.0
AI Model
Sonnet 4.5
Referenced as a benchmark comparison point on Terminal-Bench 2.0
Benchmark
Terminal-bench 2.0
Agent evaluation benchmark for terminal use maintained by the Laude Institute
Evaluation Framework
Harbor
Official evaluation framework used to compute Composer 1.5's Terminal-Bench 2.0 score
Key Actionable Insights
1Switch to Composer 1.5 or Auto model selection to take advantage of the significantly increased usage limits. Since these models draw from a separate, more generous usage pool, you can perform more agentic coding tasks without worrying about hitting API usage caps.This is especially relevant if you've been rationing API usage. The Auto + Composer pool provides substantially more headroom for daily agentic coding workflows.
2Take advantage of the temporary 6x usage promotion through February 16 to experiment with agentic coding workflows at scale. This limited-time boost lets you test Composer 1.5 extensively on real projects before the limit settles at the permanent 3x level.Use this window to evaluate whether Composer 1.5 meets your needs for large-scale codebase changes, helping you decide between the Auto + Composer pool and specific frontier models via API.
3Monitor your usage via the new in-editor usage settings page to understand your consumption patterns across both pools. This visibility helps you make informed decisions about when to use the included Auto + Composer models versus paying for specific API models.Understanding your usage split helps optimize costs — use the generous Auto + Composer pool for routine agentic tasks and reserve API usage for cases requiring specific frontier models.
4Consider that Composer 1.5 scores above Sonnet 4.5 on Terminal-Bench 2.0 while consuming from the more generous usage pool. For many agentic coding tasks, this model may be the better choice in terms of both capability and cost-effectiveness.The benchmark comparison helps inform model selection — Composer 1.5 is competitive for agent-based terminal workflows while being more sustainable for daily use.
Common Pitfalls
1
Assuming all model usage draws from the same pool. Cursor now separates Auto + Composer usage from API-priced model usage. Selecting a specific frontier model via the API pool consumes from your $20+ monthly API allowance, while Auto and Composer 1.5 use a separate, more generous pool.
Check which pool your selected model uses before starting intensive agentic coding sessions to avoid unexpected usage limits.
2
Missing the temporary 6x usage promotion that expires February 16. The permanent Composer 1.5 limit is 3x that of Composer 1, but the promotional period doubles that to 6x. Developers who don't try Composer 1.5 during this window miss an opportunity to evaluate the model with extra headroom.
The promotional period is a limited-time offer designed to encourage adoption of the new model while usage patterns stabilize.
3
Assuming Composer 1.5 is the best model for all tasks because it beats Sonnet 4.5 on one benchmark. The article explicitly notes Composer 1.5 scores below the best frontier models, meaning there are tasks where paying for API access to top-tier models will produce better results.
Use Composer 1.5 for its cost-effectiveness on routine agentic tasks, but consider frontier models via the API pool for the most demanding coding challenges.
Related Concepts
Agentic Coding
AI Code Assistants
Llm Benchmarking
Terminal-bench 2.0
Usage-based Pricing Models
Frontier AI Models
Model Evaluation Frameworks
Ai-assisted Software Development
Code Generation Models