Credit Costs

Understand how tools in Jinba Flow are charged to your organization

The usage page. Track Jinba Credit and Copilot Credit consumption by workflow, tool, and user

Overview

Jinba Flow charges two balances: Jinba Credits, which pay for running tools on Jinba’s API keys, and Copilot Credits, which pay for AI assistance. They are used for different things and reset on different schedules.

Note: Credits and subscription tiers are managed at the organization level. All members of an organization share the same subscription plan and pooled credits.

At a Glance

• Some tools are charged to your organization’s balance when they run on Jinba’s API keys.
• For tools that support it, you can bring your own API key. The provider then bills you directly and nothing is charged to your balance.
• The app shows your balance, your usage and the price of each step in US dollars.
• A tool’s configuration page shows a badge: a fixed price, or “Dynamic pricing” for a tool whose price depends on the run.
• The usage page shows what each run was actually charged.

Credit Types

Jinba Credits

Jinba Credits pay for executing tools in workflows. They let you use Jinba’s API keys without needing your own credentials. Credits are pooled per organization.

• Reset Period: Monthly (resets on subscription renewal or every 30 days for enterprise users)
• Usage: Deducted when running workflow tools like AI models, search APIs, and data processing services
• Pricing: Each tool is priced in one of the ways described in How Tools Are Priced
• Custom API Key Bypass: You can provide your own API keys to bypass charges for certain tools
• Monthly usage included:
  • Free: $3
  • Standard: $20
  • Professional: $250
  • Enterprise: Custom

A Free organization that still shows $2.50 moves to $3 at its next monthly reset.

The Standard plan is no longer sold: Professional is the only paid plan you can subscribe to. Organizations already on Standard keep it.

Copilot Credits

Copilot Credits pay for AI assistance features and copilot functionality, such as generating step descriptions, summaries, and metadata. Credits are pooled per organization.

• Reset Period: Daily (resets every day at midnight)
• Usage: Deducted when using copilot features (1 credit per request)
• Daily Copilot Requests:
  • Free: 100
  • Standard: 300
  • Professional: 500
  • Enterprise: Custom

In some organizations, including Free plan organizations, Copilot chat does not use the daily Copilot Credits. Each turn is charged by usage from the organization’s Jinba Credit balance instead, and is listed as Copilot in the usage breakdown. A turn needs a set amount of balance to start. A turn that costs more than that amount is still charged in full, so the balance can drop below $0. The next turn is then refused until the balance is topped up or reset.

How Tools Are Priced

Tools are priced in one of four ways:

• Per token: for the LLM tools
• From the step’s settings: for image and video generation, and Browser Use Cloud
• From the provider’s charge: for usage-based search
• A fixed price per run: for everything else

Prices can differ between organizations. The usage page shows what your organization was actually charged.

Per-Token Pricing (LLM Tools)

These tools are billed from the tokens a run actually uses:

• OPENAI_INVOKE / OPENAI_INVOKE_WITH_FILE
• ANTHROPIC_INVOKE / ANTHROPIC_INVOKE_WITH_FILE
• GEMINI_INVOKE / GEMINI_INVOKE_WITH_FILE / GEMINI_INVOKE_WITH_IMAGE
• GROK_CHAT
• OPENAI_DEEP_RESEARCH
• WORD_TEMPLATE_GENERATE
• TRANSLATE_PPTX_FILE
• LLAMA_CLOUD_QUERY
• ANTHROPIC_AGENT
• BROWSER_USE_SANDBOX_RUN_TASK

How it works:

1. What is counted. The cost is calculated from the model you select and the tokens the run reports:

   • input and output tokens
   • on models that support prompt caching, cached reads and cache writes, each at its own rate
   • with web search enabled, each search the provider runs

2. Long-context rates. Some models bill a whole request at higher rates once its prompt (input plus cached tokens) crosses a long-context threshold, and the cost follows those rates:

   • the gpt-5.5 and gpt-5.6 families above 272K tokens
   • gemini-3.1-pro-preview and gemini-2.5-pro above 200K
   • the Grok models from 200K

   ANTHROPIC_AGENT and BROWSER_USE_SANDBOX_RUN_TASK report tokens summed over many requests, so they are always priced at the base rate.

3. Reservation. Before the step runs, a reservation is held against your balance, so a run cannot start unless you can cover it. When the step finishes, the reservation is replaced by the real cost, which is usually lower.

4. Runs without usage. When a run reports no token usage at all, or the model it ran has no published rate, it keeps its reservation. Each tool also has a minimum charge, and such a run is never charged less than that. BROWSER_USE_SANDBOX_RUN_TASK is the exception: it settles at its minimum, not its reservation. A run that does report tokens on a priced model settles at its actual token cost, with no minimum applied.

Some tools reserve much more than their minimum, because one run can use a great deal more:

Tool	What the reservation depends on
OPENAI_DEEP_RESEARCH	The model: o3-deep-research reserves several times what o4-mini-deep-research (the default) does
BROWSER_USE_SANDBOX_RUN_TASK	max_steps and llm_model. Higher max_steps reserves more, up to the 100-step maximum
WORD_TEMPLATE_GENERATE	The tool’s own 16,000-token output limit
GROK_CHAT	The model: grok-4.5 reserves more than the default grok-4.3
TRANSLATE_PPTX_FILE	Sized for about a 20-slide deck. A larger deck can cost more than the reservation
LLAMA_CLOUD_QUERY	A fixed token estimate for one query

Every other tool in the list above reserves close to its minimum.

Steps inside a subflow or loop step are charged too. Before it starts, the enclosing step reserves one pass over its inner steps. It then settles at what the inner steps that actually ran cost, as a single entry on the enclosing step rather than one per inner step.

A run that cannot cover its reservation fails to start with an insufficient-balance error. This happens even when the work itself would have cost far less than the amount held.

The balance a run can draw on is your organization’s balance minus what its other unfinished runs have reserved, so runs started together each need their own headroom.

Settlement never checks the balance. A run that costs more than it reserved can leave the balance negative, and until it is topped up no new run that reserves credits can start.

Media Generation Pricing

These tools produce no tokens, so they are priced from what they generate. The cost depends only on the step’s settings.

• OPENAI_SORA2 is billed per second of video. The rate is set by the model (sora-2 or sora-2-pro) and, for sora-2-pro, the output resolution (720x1280 / 1280x720, or 1024x1792 / 1792x1024).
• GEMINI_GENERATE_IMAGE is billed per image generated, at a rate set by the model (gemini-3.1-flash-image, gemini-nano-banana-2.1, gemini-3.1-flash-lite-image, gemini-2.5-flash-image or gemini-3-pro-image). New steps default to gemini-3.1-flash-image.

A Gemini image step keeps running on the model it names until Google shuts that model down, and is charged at that model’s rate. Once the model is retired, the step runs on its successor and is charged the successor’s rate instead. gemini-2.5-flash-image, for example, moves to gemini-3.1-flash-lite-image on 2027-03-15.

Sometimes a setting that drives the cost (a model, a duration, a resolution) is filled in by an expression such as {{ steps.plan.output.model }}. Its value is not known until the step runs, so the reservation assumes the most expensive option it could resolve to. The final charge is the price of what actually ran.

Usage-Based Search

These tools are charged from what the provider bills for each call (can use your own API key):

• PERPLEXITY_SEARCH - Tokens at the selected Sonar model’s rate plus Perplexity’s per-request fee. Sonar Deep Research also bills its searches, citation tokens and reasoning tokens, so it costs much more than sonar. A run that reports no usage keeps its reservation.
• EXAAI_SEARCH - Priced by search type, number of results and contents (text, highlights). More results, deeper search types and more contents cost more.

Browser Use Cloud

• BROWSER_USE_RUN_TASK - Priced from max_steps: Browser Use’s per-task fee plus its per-step fee, so a larger max_steps costs more. The step is charged for its full max_steps even when the task finishes in fewer steps (can use your own API key).

Fixed Price Per Run

These tools are charged a fixed price each time they run:

• PYTHON_SANDBOX_RUN - Execute Python code in a secure sandbox (always requires credits)
• SERPAPI_SEARCH - Search using SerpAPI (can use your own API key)
• FIRECRAWL_SCRAPE - Scrape web content using Firecrawl (can use your own API key)
• ARXIV_INVOKE_AGENT - Access research papers from arXiv (always requires credits)
• YOUTUBE_GET_TRANSCRIPT - Extract transcripts from YouTube videos (always requires credits)
• LLAMA_CLOUD_PARSE - Parse documents with LlamaCloud (can use your own API key)
• JINBA_FLEX_PARSER - Flexible document parsing (can use your own API key)
• JINBA_MODULES_PARSE, JINBA_MODULES_EXTRACT, JINBA_MODULES_CHECKER_V2, CHECKER_CHECK_BY_JSON, CHECKLIST - Run on Jinba’s model keys

Knowledge Base Uploads

Adding a file to a knowledge base is charged to your Jinba Credits from the tokens the upload uses. Two kinds of tokens are counted:

• the embedding tokens of every chunk (text-embedding-3-large)
• for PDF pages that have no text layer, the Gemini OCR tokens that read them

Each file is recorded as one charge. A file cannot be added while the organization has no Jinba Credits left. See Knowledge.

Copilot Credit Usage

Copilot Credits are consumed when using AI assistance features:

• Step Description Generation: 1 credit per request
• Flow Summary Generation: 1 credit per request
• Metadata Generation: 1 credit per request
• Cron Expression Parsing: 1 credit per request
• Chat/Copilot Conversations: 1 credit per request, or charged from the Jinba Credit balance (see the note under Copilot Credits)

Bring Your Own API Key

For tools that support it, you can provide your own API keys to avoid being charged:

When a tool supports “bring your own API key,” you can configure your own credentials to bypass charges. Tools marked “always requires credits” are charged regardless of configuration.

How to Configure Your API Keys

1. Navigate to the Credentials section in your workspace settings
2. Select the appropriate service (e.g., Anthropic, OpenAI, Gemini)
3. Enter your API key
4. Save your configuration

Once configured, the system uses your API key instead of charging your balance when you use those tools.

A step whose API-key field points at your own secret only ever runs on that key. If the secret it references has been deleted or is empty, the step fails with a MissingApiKeySecret error before the tool is called; it does not silently run on Jinba’s key. Re-create the secret, or set the field back to Jinba credits.

Tools That Always Require Credits

Some tools always require credits because they use Jinba’s infrastructure or specialized services:

• ARXIV_INVOKE_AGENT - Uses OpenAI API + E2B sandbox
• YOUTUBE_GET_TRANSCRIPT - Uses Smartproxy residential proxy
• PYTHON_SANDBOX_RUN - Uses E2B code interpreter sandbox

These tools cannot use external API keys and are always charged to your balance.