OpenRouterLLMNode - ComfyUI Built-in Node Documentation

The OpenRouter LLM node sends a text prompt to a curated set of popular language models available through the OpenRouter service and returns the generated text response. It supports models from Anthropic (Claude), OpenAI (GPT), Google (Gemini), xAI (Grok), DeepSeek, Qwen, Mistral, Z.AI (GLM), Moonshot (Kimi), and Perplexity Sonar, and can optionally include images or videos as reference inputs in the request.

Inputs

When a model is selected in the model selector, the node shows model-specific widgets on top of the common inputs — reasoning effort, web search size, and/or reference media slots — depending on the capabilities of the chosen model.

Common Inputs

Parameter	Description	Data Type	Required	Range
`prompt`	Text input to the model.	STRING	Yes	N/A
`model`	The OpenRouter model used to generate the response.	DYNAMIC_COMBO	Yes	Multiple options available (see model sections below)
`seed`	Seed for sampling. Set to 0 to omit. Most models treat this as a hint only. (default: 0)	INT	Yes	0 to 2147483647
`system_prompt`	Foundational instructions that dictate the model’s behavior. (default: "")	STRING	No	N/A

Anthropic Claude Models Inputs

Shared by anthropic/claude-opus-5, anthropic/claude-opus-4.8, anthropic/claude-opus-4.7, anthropic/claude-fable-5, anthropic/claude-sonnet-5, and anthropic/claude-haiku-4.5.

Parameter	Description	Data Type	Required	Range
`reasoning_effort`	Reasoning effort. ‘off’ disables reasoning entirely. (default: “off”)	COMBO	No	”off" "low" "medium" "high”

These models support up to 20 reference images (see Reference Inputs).

OpenAI GPT Models Inputs

Shared by openai/gpt-5.6-sol-pro, openai/gpt-5.6-sol, openai/gpt-5.6-terra-pro, openai/gpt-5.6-terra, openai/gpt-5.6-luna-pro, openai/gpt-5.6-luna, openai/gpt-5.5-pro, and openai/gpt-5.5.

Parameter	Description	Data Type	Required	Range
`reasoning_effort`	Reasoning effort. ‘off’ disables reasoning entirely. (default: “off”)	COMBO	No	”off" "low" "medium" "high”

These models support up to 20 reference images (see Reference Inputs).

Google Gemini 3.5 Flash Inputs

Parameter	Description	Data Type	Required	Range
`reasoning_effort`	Reasoning effort. ‘off’ disables reasoning entirely. (default: “off”)	COMBO	No	”off" "low" "medium" "high”

This model supports up to 20 reference images and up to 4 reference videos (see Reference Inputs).

xAI Grok Models Inputs

Shared by x-ai/grok-4.5, x-ai/grok-4.20, and x-ai/grok-4.3.

Parameter	Description	Data Type	Required	Range
`reasoning_effort`	Reasoning effort. ‘off’ disables reasoning entirely. (default: “off”)	COMBO	No	”off" "low" "medium" "high”

These models support up to 20 reference images (see Reference Inputs).

DeepSeek Models Inputs

Shared by deepseek/deepseek-v4-pro, deepseek/deepseek-v4-flash, and deepseek/deepseek-v3.2.

Parameter	Description	Data Type	Required	Range
`reasoning_effort`	Reasoning effort. ‘off’ disables reasoning entirely. (default: “off”)	COMBO	No	”off" "low" "medium" "high”

Text-only models — no reference images or videos.

Qwen 3.6 Max Preview Inputs

Parameter	Description	Data Type	Required	Range
`reasoning_effort`	Reasoning effort. ‘off’ disables reasoning entirely. (default: “off”)	COMBO	No	”off" "low" "medium" "high”

Text-only model — no reference images or videos.

Qwen 3.6 Plus and Qwen 3.6 Flash Inputs

Shared by qwen/qwen3.6-plus and qwen/qwen3.6-flash.

Parameter	Description	Data Type	Required	Range
`reasoning_effort`	Reasoning effort. ‘off’ disables reasoning entirely. (default: “off”)	COMBO	No	”off" "low" "medium" "high”

These models support up to 10 reference images and up to 4 reference videos (see Reference Inputs).

Mistral Large 2512 Inputs

No profile-specific inputs (standard profile). This model supports up to 8 reference images (see Reference Inputs).

Mistral Medium 3.5 Inputs

Parameter	Description	Data Type	Required	Range
`reasoning_effort`	Reasoning effort. ‘off’ disables reasoning entirely. (default: “off”)	COMBO	No	”off" "low" "medium" "high”

This model supports up to 8 reference images (see Reference Inputs).

Z.AI GLM Models Inputs

Shared by z-ai/glm-4.6 and z-ai/glm-5.

Parameter	Description	Data Type	Required	Range
`reasoning_effort`	Reasoning effort. ‘off’ disables reasoning entirely. (default: “off”)	COMBO	No	”off" "low" "medium" "high”

Text-only models — no reference images or videos.

Moonshot Kimi K3 and K2.6 Inputs

Shared by moonshotai/kimi-k3 and moonshotai/kimi-k2.6.

Parameter	Description	Data Type	Required	Range
`reasoning_effort`	Reasoning effort. ‘off’ disables reasoning entirely. (default: “off”)	COMBO	No	”off" "low" "medium" "high”

These models support up to 10 reference images (see Reference Inputs).

Moonshot Kimi K2 Thinking Inputs

Parameter	Description	Data Type	Required	Range
`reasoning_effort`	Reasoning effort. ‘off’ disables reasoning entirely. (default: “off”)	COMBO	No	”off" "low" "medium" "high”

Text-only model — no reference images or videos.

Perplexity Sonar Pro Inputs

Parameter	Description	Data Type	Required	Range
`search_context_size`	How much web search context to retrieve. Larger = more grounded but slower/pricier. (default: “medium”)	COMBO	No	”low" "medium" "high”

Text-only model — no reference images or videos.

Perplexity Sonar Reasoning Pro and Sonar Deep Research Inputs

Shared by perplexity/sonar-reasoning-pro and perplexity/sonar-deep-research.

Parameter	Description	Data Type	Required	Range
`search_context_size`	How much web search context to retrieve. Larger = more grounded but slower/pricier. (default: “medium”)	COMBO	No	”low" "medium" "high”
`reasoning_effort`	Reasoning effort. ‘off’ disables reasoning entirely. (default: “off”)	COMBO	No	”off" "low" "medium" "high”

Text-only models — no reference images or videos.

Reference Inputs

Parameter	Description	Data Type	Required	Range
`images`	Optional reference image(s), sent as URLs. Growable slot: connect 1..N image inputs (`image_1`, `image_2`, …); the count limit depends on the selected model (see model sections).	IMAGE	No	0 to 20 (model-dependent: 8, 10, or 20)
`videos`	Optional reference video(s), sent as URLs. Growable slot: connect 1..N video inputs (`video_1`, `video_2`, …); the count limit depends on the selected model (see model sections).	VIDEO	No	0 to 4 (model-dependent)

Notes:

Available models: The available model options are dynamically built and include models with different capabilities. The full list of 34 models is:
- Anthropic: anthropic/claude-opus-5, anthropic/claude-opus-4.8, anthropic/claude-opus-4.7, anthropic/claude-fable-5, anthropic/claude-sonnet-5, anthropic/claude-haiku-4.5
- OpenAI: openai/gpt-5.6-sol-pro, openai/gpt-5.6-sol, openai/gpt-5.6-terra-pro, openai/gpt-5.6-terra, openai/gpt-5.6-luna-pro, openai/gpt-5.6-luna, openai/gpt-5.5-pro, openai/gpt-5.5
- Google: google/gemini-3.5-flash
- xAI: x-ai/grok-4.5, x-ai/grok-4.20, x-ai/grok-4.3
- DeepSeek: deepseek/deepseek-v4-pro, deepseek/deepseek-v4-flash, deepseek/deepseek-v3.2
- Qwen: qwen/qwen3.6-max-preview, qwen/qwen3.6-plus, qwen/qwen3.6-flash
- Mistral: mistralai/mistral-large-2512, mistralai/mistral-medium-3-5
- Z.AI: z-ai/glm-4.6, z-ai/glm-5
- Moonshot: moonshotai/kimi-k3, moonshotai/kimi-k2.6, moonshotai/kimi-k2-thinking
- Perplexity: perplexity/sonar-pro, perplexity/sonar-reasoning-pro, perplexity/sonar-deep-research
Image and video constraints: The maximum number of reference images and videos depends on the selected model. The node raises an error if the total number of images or videos provided exceeds the model’s limit. Models without image or video support do not show the corresponding reference slots.
Reasoning behavior: When reasoning_effort is set to anything other than “off”, the request asks the provider to reason internally without returning the reasoning trace.
Seed behavior: The seed parameter has a “control_after_generate” behavior, meaning it can be set to automatically change (e.g., randomize, increment, or fixed) after each node execution, depending on the user’s widget settings.
System prompt: The system_prompt parameter is optional and is marked as an advanced parameter in the user interface.
Error cases: The node raises an error if the prompt is empty after trimming whitespace, if OpenRouter returns an error, if the selected model refuses to respond, or if the response contains no choices or message. A price badge on the node shows an approximate cost estimate per 1K tokens based on the selected model.

Outputs

Output Name	Description	Data Type
`output`	The generated text response from the OpenRouter model.	STRING

This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub

Source fingerprint (SHA-256): 534ab9ecc12e35a23a4d8f3e10f4f82d95db8e902ac8a2f2ee0ea68246516f62

​Inputs

​Common Inputs

​Anthropic Claude Models Inputs

​OpenAI GPT Models Inputs

​Google Gemini 3.5 Flash Inputs

​xAI Grok Models Inputs

​DeepSeek Models Inputs

​Qwen 3.6 Max Preview Inputs

​Qwen 3.6 Plus and Qwen 3.6 Flash Inputs

​Mistral Large 2512 Inputs

​Mistral Medium 3.5 Inputs

​Z.AI GLM Models Inputs

​Moonshot Kimi K3 and K2.6 Inputs

​Moonshot Kimi K2 Thinking Inputs

​Perplexity Sonar Pro Inputs

​Perplexity Sonar Reasoning Pro and Sonar Deep Research Inputs

​Reference Inputs

​Outputs

Inputs

Common Inputs

Anthropic Claude Models Inputs

OpenAI GPT Models Inputs

Google Gemini 3.5 Flash Inputs

xAI Grok Models Inputs

DeepSeek Models Inputs

Qwen 3.6 Max Preview Inputs

Qwen 3.6 Plus and Qwen 3.6 Flash Inputs

Mistral Large 2512 Inputs

Mistral Medium 3.5 Inputs

Z.AI GLM Models Inputs

Moonshot Kimi K3 and K2.6 Inputs

Moonshot Kimi K2 Thinking Inputs

Perplexity Sonar Pro Inputs

Perplexity Sonar Reasoning Pro and Sonar Deep Research Inputs

Reference Inputs

Outputs