Skip to content
Model infoModelsSnapshot: 15 Oct 2026

Cursor / Composer

Composer 2

Kimi K2.5-derived agentic coding model with continued pretraining, full-parameter RL and self-summarization.

restrictedClosed weightsCoding
Official documentation (opens in a new tab)Reviewed 15 Oct 2026
10 of 19 specification fields collected
Context windowNot collectedAwaiting source review
Maximum outputNot collectedAwaiting source review
Input billingSee ratesCursor Composer 2 · standard
Output billingSee ratesPublished units and conditions below

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit
Not collected

Exact version-specific model contract not established. Product capabilities do not automatically establish a standalone inference API feature.

Maximum input tokens
Not collected

Exact version-specific model contract not established. Product capabilities do not automatically establish a standalone inference API feature.

Maximum output
Not collected

Exact version-specific model contract not established. Product capabilities do not automatically establish a standalone inference API feature.

Parameters
1.04T total; 32B activeSource for Parameters (opens in a new tab)
Tool calling
SupportedSource for Tool calling (opens in a new tab)

Trained for the Cursor agent tool harness. Not a claim of arbitrary OpenAI-compatible function-call schemas.

Structured output
Not collected

Exact version-specific model contract not established. Product capabilities do not automatically establish a standalone inference API feature.

Streaming
Not collected

Exact version-specific model contract not established. Product capabilities do not automatically establish a standalone inference API feature.

Batch processing
Not collected

Exact version-specific model contract not established. Product capabilities do not automatically establish a standalone inference API feature.

Fine-tuning
Not collected

Exact version-specific model contract not established. Product capabilities do not automatically establish a standalone inference API feature.

Context caching
Not collected

Exact version-specific model contract not established. Product capabilities do not automatically establish a standalone inference API feature.

Knowledge cutoff
Not collected

Exact version-specific model contract not established. Product capabilities do not automatically establish a standalone inference API feature.

Input modalities
TextSource for Input modalities (opens in a new tab)

Native image support is not established here; agent tools can handle media without proving model-native generation/vision.

Base model
Moonshot Kimi K2.5Source for Base model (opens in a new tab)
Architecture
Mixture of experts, with added multi-token prediction layers and speculative decoding.Source for Architecture (opens in a new tab)
Published training method
Continued pretraining, targeted SFT and full-parameter asynchronous RL; MXFP8 pretraining and NVFP4-aware RL inference.Source for Published training method (opens in a new tab)
Training context length
32k initial sequence length, then a 256k long-context training phase.Source for Training context length (opens in a new tab)

Research training sequence lengths are not a published current product input limit.

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

Cursor Composer 2 · standard

Provider source (opens in a new tab)
API model ID
composer-2

Cloud Agents example model ID, not a standalone chat-completions model.

Access
Historical launch tier. Current Cloud Agents examples show this family; exact present availability/rates require account model discovery.
Regions
Default model-specific processing region not established. Enterprise US-only residency can cover Composer when configured; external integrations have separate boundaries.
API parameters
Use model IDs/parameters returned by GET /v1/models; do not assume old defaults are unchanged.Source for API parameters (opens in a new tab)
API endpoint URL
POST https://api.cursor.com/v1/agents; discover supported models and parameters with GET /v1/models.Source for API endpoint URL (opens in a new tab)

Cloud agent workflow API, not a standalone model-inference or chat-completions endpoint.

Authentication
User or service-account API key. HTTP Basic with key as username and empty password, or Bearer token.Source for Authentication (opens in a new tab)
Output formats
Agent run/task state, messages and artifacts. The API includes an event-stream route; that does not establish a standalone model token-stream interface.Source for Output formats (opens in a new tab)

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 15 Oct 2026
CategoryUsageRateUnit
Historical launch pricingInput tokens1$0.501M tokens
Historical launch pricingOutput tokens1$2.501M tokens

Rate conditions

  1. March 19, 2026 launch price, not verified as the current rate. Current question estimates must not use this historical observation.
Pricing source (opens in a new tab)

Data & enterprise terms

With Privacy Mode enabled, code is not used for training by Cursor or model providers. Custom-model inference providers may process it. Cloud Agents store encrypted repository copies during execution, with deletion after completion described by the provider. These claims do not establish unconditional zero retention for every feature or account.

Read the service policy (opens in a new tab)

Cursor Composer 2 · fast

Provider source (opens in a new tab)
API model ID
composer-2

Cloud Agents example model ID, not a standalone chat-completions model.

Access
Historical launch tier. Current Cloud Agents examples show this family; exact present availability/rates require account model discovery.
Regions
Default model-specific processing region not established. Enterprise US-only residency can cover Composer when configured; external integrations have separate boundaries.
API parameters
The API example selects model.id composer-2 with params [{id: fast, value: true}]. Verify supported combinations with GET /v1/models.Source for API parameters (opens in a new tab)
API endpoint URL
POST https://api.cursor.com/v1/agents; discover supported models and parameters with GET /v1/models.Source for API endpoint URL (opens in a new tab)

Cloud agent workflow API, not a standalone model-inference or chat-completions endpoint.

Authentication
User or service-account API key. HTTP Basic with key as username and empty password, or Bearer token.Source for Authentication (opens in a new tab)
Output formats
Agent run/task state, messages and artifacts. The API includes an event-stream route; that does not establish a standalone model token-stream interface.Source for Output formats (opens in a new tab)

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 15 Oct 2026
CategoryUsageRateUnit
Historical launch pricingInput tokens1$1.501M tokens
Historical launch pricingOutput tokens1$7.501M tokens

Rate conditions

  1. March 19, 2026 launch price, not verified as the current rate. Current question estimates must not use this historical observation.
Pricing source (opens in a new tab)

Data & enterprise terms

With Privacy Mode enabled, code is not used for training by Cursor or model providers. Custom-model inference providers may process it. Cloud Agents store encrypted repository copies during execution, with deletion after completion described by the provider. These claims do not establish unconditional zero retention for every feature or account.

Read the service policy (opens in a new tab)

Caching

Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.

01 / INITIAL INPUTProcess & retain

The system processes eligible input and retains reusable state.

02 / LATER REQUESTCheck available state

Reuse depends on matching input, session state and the cache policy.

03 / CACHE HITReuse & generate

Reusing state can reduce repeated work. Any charges follow the billing rules below.

Cursor Composer 2 · standard

composer-2
Not verified

No exact-version prompt-cache contract established. Internal KV reuse, code indexing and self-summarization are not a published cache discount.

How it works
Not collected

No exact-version prompt-cache contract established. Internal KV reuse, code indexing and self-summarization are not a published cache discount.

Minimum reusable input
Not collected

No exact-version prompt-cache contract established. Internal KV reuse, code indexing and self-summarization are not a published cache discount.

Lifetime & refresh
Not collected

No exact-version prompt-cache contract established. Internal KV reuse, code indexing and self-summarization are not a published cache discount.

How to enable it
Not collected

No exact-version prompt-cache contract established. Internal KV reuse, code indexing and self-summarization are not a published cache discount.

Usage fields
Not collected

No exact-version prompt-cache contract established. Internal KV reuse, code indexing and self-summarization are not a published cache discount.

Billing rules
Not collected

No exact-version prompt-cache contract established. Internal KV reuse, code indexing and self-summarization are not a published cache discount.

Limits & data boundaries
Not collected

No exact-version prompt-cache contract established. Internal KV reuse, code indexing and self-summarization are not a published cache discount.

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Caching reviewed 15 Oct 2026

Cursor Composer 2 · fast

composer-2
Not verified

No exact-version prompt-cache contract established. Internal KV reuse, code indexing and self-summarization are not a published cache discount.

How it works
Not collected

No exact-version prompt-cache contract established. Internal KV reuse, code indexing and self-summarization are not a published cache discount.

Minimum reusable input
Not collected

No exact-version prompt-cache contract established. Internal KV reuse, code indexing and self-summarization are not a published cache discount.

Lifetime & refresh
Not collected

No exact-version prompt-cache contract established. Internal KV reuse, code indexing and self-summarization are not a published cache discount.

How to enable it
Not collected

No exact-version prompt-cache contract established. Internal KV reuse, code indexing and self-summarization are not a published cache discount.

Usage fields
Not collected

No exact-version prompt-cache contract established. Internal KV reuse, code indexing and self-summarization are not a published cache discount.

Billing rules
Not collected

No exact-version prompt-cache contract established. Internal KV reuse, code indexing and self-summarization are not a published cache discount.

Limits & data boundaries
Not collected

No exact-version prompt-cache contract established. Internal KV reuse, code indexing and self-summarization are not a published cache discount.

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Caching reviewed 15 Oct 2026

What the measurements say

No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.

Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.

The record behind the model

Why it is included

Cursor describes its own post-trained coding model, distributed through its product. Composer 2 is described among proprietary models in its technical report. Private classification for other Composer generations is inferred from this product/research lineage; their exact weight license is not published in the reviewed sources. Public Kimi K2.5 base weights do not constitute a release of the Composer derivative.

Classification evidence (opens in a new tab)

Lifecycle dates

Model release announced19 Mar 2026Source (opens in a new tab)

Sources reviewed