Pricing & billing
Three subscription tiers plus pay-as-you-go prepaid credits. The same monthly subscription budget covers web UI and API usage.
MCP and API billing
Reviews started through MCP or the REST API bill exactly like web-app reviews. Each review counts as one metered run. RigorDesk deducts it from your subscription allowance first, then from prepaid credits. Creating assemblies, listing models, and reading results over MCP or REST is free and never consumes credits.
Tiers
| Tier | Monthly allowance | Max rounds per run | Subscription | Notes |
|---|---|---|---|---|
| Hobby | 2 debates per calendar month | 2 | $0 | Refinement and free-tier models. |
| Pro | 1,000 compute credits | 5 | $20 | All models, all 7 formats, and REST API. |
| Pro Max | 5,000 compute credits | 10 | $100 | Five times Pro compute plus API webhooks. |
One compute credit represents $0.01 of metered model-provider cost. Frontier models and longer formats consume more credits than smaller models and shorter runs. RigorDesk meters API usage at twice provider cost. Each debate's audit trail still shows token counts, but RigorDesk prices everything in compute credits.
New model rate reference
| Model | Standard input | Standard output | Context | Pricing note |
|---|---|---|---|---|
| GPT-6.1 Sol | $2/M tokens | $10/M tokens | 1,050,000 | Above 272,000 prompt tokens, rates increase to $4/M input and $15/M output |
| GPT-6 Luna | $0.10/M tokens | $0.50/M tokens | 1,050,000 | Above 272,000 prompt tokens, rates increase to $0.20/M input and $0.75/M output |
| Claude Sonnet 5.5 | $2/M tokens | $10/M tokens | 1,000,000 | Thinking is always enabled |
| Claude Opus 5.5 | $4/M tokens | $20/M tokens | 1,000,000 | Thinking is always enabled |
| DeepSeek V4.1 Flash | $0.0198/M tokens | $0.396/M tokens | 1,048,576 | Hobby access; text and image input |
| Gemini 3.8 Flash | $1.50/M tokens | $7.50/M tokens | 1,048,576 | Standard estimate rate; temporary 50% promotion excluded |
| Grok 4.7 | $2/M tokens | $6/M tokens | 500,000 | Above 200,000 prompt tokens, rates increase to $4/M input and $12/M output |
| MiMo V2.6 Pro | $0.435/M tokens | $0.87/M tokens | 1,048,576 | Xiaomi model; provider context limit |
| Muse Spark 1.3 | $1.25/M tokens | $4.25/M tokens | 1,048,576 | Multimodal reasoning model |
| Command A+ | $0.30/M tokens | $1.50/M tokens | 192,000 | Text and image input, tool calling, and structured outputs |
| GPT-6 Astra | $5/M tokens | $25/M tokens | 1,050,000 | OpenAI flagship model |
| Gemini 3.7 Flash | $0.75/M tokens | $3.75/M tokens | 1,048,576 | Existing standard estimate rate |
| DeepSeek V4 Pro 0813 | $0.435/M tokens | $0.87/M tokens | 1,048,576 | Text input |
| Grok 4.6 | $2/M tokens | $6/M tokens | 500,000 | Above 200,000 prompt tokens, rates increase to $4/M input and $12/M output |
Rates were checked against OpenRouter's model catalog on September 30, 2026. Gemini 3.8 Flash currently lists at a promotional $0.75/M input and $3.75/M output; RigorDesk estimates use its standard $1.50/M and $7.50/M rates. Other listed rates follow the pricing sync.
These are provider rates. RigorDesk meters API usage at twice provider cost. Compute-credit estimates also account for the number of models, debate rounds, reasoning tokens, web search, and growing transcript context. Pre-run estimates use base token rates and can understate long-context surcharges. The completed debate audit trail records actual provider-reported usage.
What counts as 1 debate
A "debate" is one execution of the orchestrator against one assembly, with one prompt. It may run 1 round or maxRounds rounds. Either way it counts as one debate for billing. The only exception is a debate that fails partway through. See "Failed debates" below.
A debate consumes:
- One model call per model per round, for example 3 models × 3 rounds = 9 calls
- Pairwise consensus evaluations, for example 3 models × 3 rounds × 3 pairs = 27 evaluations
- One chairman synthesis call at the end
The cost of the consensus evaluations and chairman is small relative to the model calls. The bulk of your spend is the model turns.
Prepaid credits
Prepaid credits do not expire. A positive balance unlocks all models and all seven debate formats in the web app. Subscription limits still control round depth, thinking, web search, document size, API access, and webhooks.
Use credits when:
- You hit your monthly cost cap and need to keep running debates
- You're between subscription renewals
- You want a hard spending limit. Turn off auto-renew, set up credits, and you cannot overspend.
- You want paid models and debate formats without a subscription
You can buy credits in the web app at Settings → Prepaid Credits.
Failed debates
If a debate fails partway through, RigorDesk logs the cost, but the failed debate does not count against your debate cap of 2 debates per month on Hobby. RigorDesk records the cost so you can see what happened, but you pay only for the model calls that actually completed.
A debate fails if:
- A model turn times out after 3 retries
- The chairman call fails after 3 retries
- The debate is cancelled by the user
- The budget is exhausted mid-debate
If the retry succeeds and the debate completes, RigorDesk charges the full cost normally.
Reading your cost breakdown
The web app's Settings → Usage card shows:
- Compute credits used for this period
- Prepaid credits remaining, when applicable
- Sessions count: number of completed debates counted toward the current allowance
Per-debate, the full session response includes:
totalCost: the dollar amounttotalTokens: the token count- Per-turn
costandcostfor input, output, reasoning, and cached tokens
For production monitoring, log totalCost per debate to your own metrics system. See production patterns for an example.
When you should upgrade
A simple rule:
- Hobby is for trying the platform. 2 free debates/month lets you see if it works for you.
- Pro is for serious use, with 1,000 compute credits each month.
- Pro Max is for production, with 5,000 compute credits each month.
- If you hit your cap, buy credits or upgrade.
The tier price is fixed. What matters is what you do with it. If a single $0.50 debate stops a bad plan from shipping, Pro has paid for itself. If you use RigorDesk for low-stakes ideation, Hobby plus occasional credits is enough.
Refunds and cancellation
- Cancel anytime in the web app's Billing section. Your subscription stays active until the end of the current billing period.
- No refunds for partial months. You forfeit the remaining cost cap if you cancel.
- You cannot refund prepaid credits, but they don't expire.
- See the refund policy for the full terms.
Tax and invoicing
Paddle, our payment processor, issues invoices to the email on your account. To get a VAT-compliant invoice with your company's details, update your billing info in the Customer Portal under Settings → Billing → Launch Customer Portal.
For enterprise contracts with custom terms, contact us.
