Skip to main content
This page details the pricing structure and how credits are deducted for various API operations.
To upgrade your plan, visit the Account Dashboard, open the Billing tab, and select your tier.

API Subscriptions

Flexible subscription plans that scale with your needs. Choose a plan for higher limits and features; usage is billed per operation.

Plans at a glance

Prices shown are subscription fees; API usage is billed per operation (see Usage-based pricing below). Rate limits shown are those returned by GET /v1/rate/limits for any account with an active premium subscription (pro, ultra, or premium tier) — see Rate Limits for how free vs. premium tiers are resolved.
Some plan-comparison items you may see elsewhere (e.g. SOC 2 certification, dedicated infrastructure, or request-priority routing) are account/infrastructure-level facts that could not be verified against the Swarms API application source and are intentionally omitted from this table. See Security and Priority Processing for what is verifiable.

Update your plan in 3 steps

  1. Go to the Account Dashboard.
  2. Open the Billing tab and scroll down to the Plans section.
  3. Select your desired plan. You’ll be redirected to Stripe to complete payment.

Free

Get started with AI. No monthly fees — pay only for usage.
  • Sign-up bonus credits (shown as free_credit; they don’t pay for API usage, see Checking Your Credit Balance)
  • Basic API access
  • Pay-per-use pricing
  • Community support
  • Standard processing speed
  • Access to the Marketplace
  • Execution history: your 50 newest logs from the last 7 days at /v1/account/logs
  • Note: Premium endpoints (batch processing, reasoning agents, advanced workflows) are not available on the free tier

Pro

Most popular. Perfect for professionals who need more power and features.
  • $19.99/month (save 15% with annual billing)
  • Higher rate limits: 2,000 req/min, 10,000 req/hour, 100,000 req/day, 2M tokens per agent (vs. free tier’s 100/350/1,200 and 200K — see Rate Limits)
  • Access to premium-only models (see below)
  • Execution history: your 5,000 newest logs from the last 90 days
  • Access to Premium Endpoints: Unlock batch processing, reasoning agents, and advanced workflows
    • /v1/agent/batch/completions - High-throughput batch agent processing
    • /v1/swarm/batch/completions - Batch swarm completions
    • /v1/reasoning-agent/completions - Advanced reasoning capabilities
    • /v1/batched-grid-workflow/completions - Grid workflow execution
    • Plus graph workflows
    • See full list: Premium Endpoints

Ultra

Best for growing teams and production workloads that need higher limits and security.
  • $99/month (save 15% with annual billing)
  • Everything in Pro. Note: per api/schemas.py (RateLimitsPremiumTier), Pro and Ultra currently resolve to the same numeric rate limits and per-agent token limit — any account with an active subscription (pro, ultra, or premium tier) gets the same premium allowances.
  • Full access to Premium Endpoints
  • Your full execution history, with no row or age limit (Premium is the same)
  • For compliance certifications and enhanced-security specifics, see Security

Premium-Only Models

Free tier accounts are blocked from requesting the following models on any endpoint (including /v1/agent/completions and /v1/swarm/completions), regardless of available credits. Any active Pro or Ultra subscription unlocks all of them — there is no separate model tier between Pro and Ultra:
  • gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, gpt-5.6
  • claude-fable-5, claude-fable-5-1
  • claude-opus-5, claude-opus-5-5, claude-opus-4-8
  • gemini-3.1-pro-preview, gemini-3.1-pro-preview-customtools
  • xai/grok-4.5, xai/grok-4.5-latest
  • The OpenRouter ids openrouter/openai/gpt-5.6-sol, openrouter/openai/gpt-5.6-terra, openrouter/openai/gpt-5.6-luna, openrouter/anthropic/claude-fable-5, openrouter/anthropic/claude-fable-5.1, openrouter/anthropic/claude-opus-5, openrouter/anthropic/claude-opus-5.5, openrouter/anthropic/claude-opus-4.8, openrouter/google/gemini-3.1-pro-preview, openrouter/google/gemini-3.1-pro-preview-customtools and openrouter/x-ai/grok-4.5
  • Any groq/* model
The check also matches a listed id behind another provider prefix or with a :variant suffix, e.g. deepinfra/anthropic/claude-fable-5 or openrouter/openai/gpt-5.6-sol:batch. Requesting one of these models on the free tier returns a 403 error with suggested free-tier alternatives. GET /v1/models/available leaves them out of a free-tier key’s list.

Enterprise

Critical support, compliance, and control for large-scale enterprises.
  • Contact sales for custom pricing
  • Dedicated 24/7 support
  • Custom solutions engineering
  • Onsite training and onboarding
  • Custom agent development
  • No rate limits
  • Access to experimental features

Usage-based pricing

Pay only for what you use. Each model’s tokens are billed at that model’s own price, plus a 15% service fee. Some features add a flat fee on top.

Token pricing

The model price comes from: Example prices per 1M tokens at the time of writing, service fee included: Provider prices change, so look up current prices with the endpoints in Looking up model prices instead of hardcoding them. How tokens are counted:
  • Provider-reported tokens. Agent, chat, swarm, graph workflow and batched grid workflow completions bill the tokens the model provider reports for each agent, summed over every LLM call the agent made. That includes every loop, re-sent conversation history and reasoning tokens.
  • Per agent, per model. A swarm or workflow that mixes models bills each agent’s tokens at its own model’s price, then sums them.
  • Fallback models. If an agent falls back to another model (fallback_models), its tokens are priced at the model that actually answered.
  • Cached and reasoning tokens. Cached input tokens are billed at the model’s normal input price, and reasoning tokens at its output price.
  • Reasoning agents. AgentJudge and ReflexionAgent bill the provider-reported tokens of every LLM call their internal agents make, every loop included. self-consistency and consistency-agent send the prompt once per sample, so they bill the API’s count of the prompt (agent name, description, system prompt and task) × num_samples as input, plus the counted tokens of the returned outputs: num_samples: 5 costs about five times the prompt. num_samples must be 1 to 2,000 (422 otherwise), and output_type is ignored.

Flat fees

MCP servers and the web_scraper tool have no fee: only the tokens the agent uses are billed. On OpenRouter free models the tokens and the agent cost are $0, but the image and search fees still apply, and so does the Auto Agent Builder’s $0.01. A run that costs $0 deducts nothing, but it still counts toward your rate limits, and the $1.00 minimum balance is still checked.

By endpoint

Some swarm types build agents of their own: the HierarchicalSwarm director, the MajorityVoting consensus agent, the CouncilAsAJudge aggregator and the LLMCouncil chairman. Their tokens are billed too, each at its own model, but the $0.01 agent cost counts only the agents in your request. HeavySwarm builds all of its agents itself, so its tokens are priced at heavy_swarm_worker_model_name.

Discounts

Night-time discount: for paid members (tier pro, premium or ultra, or an active subscription), swarm completions get 50% off the token cost from 8 PM to 6 AM Pacific Time, judged when the run is billed, after it finishes. Free accounts pay the full token price at any hour. The $0.01 agent cost is not discounted, and no other endpoint gets the discount.

Worked examples

Prices below are illustrative provider prices before the fee: gpt-5.4 at $2.50 / $15.00 and claude-sonnet-5 at $2.00 / $10.00 per 1M input / output tokens. Mixed-model swarm, daytime. A gpt-5.4 agent with 500,000 input and 100,000 output tokens, and a claude-sonnet-5 agent with 200,000 input and 300,000 output tokens:
At night on a paid plan the two token costs are halved: $0.02 + $1.58125 + $1.955 = $3.55625. Agent completion with an image. A claude-sonnet-5 agent with 250,000 input and 100,000 output tokens and one image:
Unpriced model. 1,000,000 input and 500,000 output tokens on a model with no price data:

Looking up model prices

GET /v1/models/available lists every model available to your tier with its price per 1M tokens, service fee included:
GET /v1/models/pricing prices one model. Any model name gets a price; a name with no price data comes back with "price_source": "fallback" and the fallback rates, which is what a run on it is billed at. An OpenRouter free model comes back at $0 with "price_source": "openrouter".
These prices already include the service fee, so a run’s token cost is tokens / 1,000,000 × price. The OpenAI-compatible GET /v1/models carries the same numbers in a pricing object on each model. See Available Models.

Retrieving the pricing configuration

GET /v1/usage/costs returns the flat fees, the night-time discount, the flash sale settings and the fallback rates:
Example Response:
The fallback rates apply only to models with no price data, and are before the service fee. For a specific model’s price, use GET /v1/models/pricing. flash_sale_active says whether a flash sale is on. While it is true, tokens cost flash_sale_discount of the provider price plus the 15% service fee (0.5 + 0.15 = 0.65 of the provider price), unless the night-time discount is cheaper; the two never stack. While it is false, flash_sale_discount has no effect.

What the response reports

Every billed response reports the tokens and the amount charged in its usage block, or in usage.billing_info for swarm completions. A gpt-5.4 agent completion with 12,000 input and 4,000 output tokens:
total_cost includes the image fee and the $0.04 auto_search fee when they apply. flash_sale_applied is true only when the run was billed at the flash sale price.

Checking Your Credit Balance

Use GET /v1/account/credits to check your current balance before making billable requests:
Only purchased credit (credit) pays for API usage. free_credit and referral_credits are still returned, but they are never spent and don’t count toward the $1.00 minimum. total_credits equals credit, and it is the number checked against the $1.00 minimum for billable routes. Top up at cloud.swarms.world/settings.

Important Notes