Skip to main content
OpenRouter bills each provider’s list price and adds a 5.5% fee when you buy Standard credits by card. A new personal Capriole AI Premium subscription costs USD 9 per month and includes 5M API tokens, enough for the example below. The same models cost USD 6.75 to USD 90 at paid Standard list prices, while the calculated OpenRouter cash total is USD 7.55 to USD 94.95 after its published credit-purchase fee. Premium, the USD 9 plan, also includes unlimited web chat with the four flagships: GPT-6.2 Astra, Claude Fable 5.1, Claude Opus 5.5, and Gemini 4 Carbon. For Gemini 3.8 Flash, this workload costs USD 6.75 through the direct API or USD 7.55 through OpenRouter including its credit-purchase fee. The USD 9 Premium subscription also includes unlimited web chat. This is a reproducible comparison, not a universal price promise. It holds the token mix constant and uses the cited public prices. Rates checked November 18, 2026.

How much do 5M tokens cost on each API?

With 4M input and 1M output tokens, GPT-6 Astra and Claude Fable 5.1 cost USD 90 through the official APIs and USD 94.95 through OpenRouter, and GPT-6.2 Astra costs USD 72 and USD 75.96. If your Capriole AI token balance covers the example, you pay nothing extra. A personal Premium top-up costs USD 8 for 5M additional tokens. The table holds billed token counts constant, not source text or semantic workload. Tokenizers differ between model families. The scenario uses 100 requests with 40,000 uncached input tokens and 10,000 output tokens each, keeping every prompt below OpenAI’s and Google’s long-prompt price thresholds. The Claude, OpenAI, and Gemini API pricing articles use three input tokens per output token instead, so their 5M-token totals are higher, such as USD 100 for GPT-6 Astra. The scenario excludes cache reads and writes, tools, free tiers, alternative service tiers, batch processing, taxes, and account-specific discounts. * Premium costs USD 9 per month and includes unlimited browser chat across the supported Premium model set plus the first 5 million monthly charged tokens. This is the new customer’s public purchase price for that complete workspace bundle, not a per-model token rate or marginal inference cost. An existing Premium member with enough included balance pays no additional cash for this usage. After the included allowance is used, one additional 5 million charged-token pack has a USD 8 base price. Personal top-ups require active personal Premium access. Gemini 4 Carbon and Gemini 4 Argon use Google’s prices through November 30, 2026. From December 1, both cost USD 4 / USD 20, which makes this workload USD 36 direct and USD 37.98 through OpenRouter with the card fee. Try Capriole AI in free browser chat before you subscribe to Premium for the full workspace and API access.

How to reproduce the calculation

Direct providers publish separate prices for input and output. For this workload, the formula is:
GPT-6 Astra, Claude Fable 5.1, and Claude Fable 5 therefore cost:
OpenRouter states that it passes through the selected provider’s inference price without a markup. Its Standard plan lists a 5.5% fee, with a USD 0.80 minimum, when you buy credits by card. At the displayed inference rate, the cash calculation is:
The same arithmetic produces USD 75.96 for GPT-6.2 Astra (4 × USD 8 + 1 × USD 40 = USD 72 direct), USD 37.98 for Claude Opus 5.5, USD 18.99 for Gemini 4 Carbon, Gemini 4 Argon, and GPT-6.2 Sol, and USD 21.10 for Gemini 3.1 Pro Preview. Gemini 3.8 Flash costs USD 6.75 at Google’s introductory Standard rates through December 31, 2026 (4 × USD 0.75 + 1 × USD 3.75). OpenRouter’s USD 0.80 minimum card fee brings the total to USD 7.55. For Gemini 3.1 Pro Preview, the input and output rates per 1M tokens give a direct cost of 4 × USD 2 + 1 × USD 12 = USD 20. On OpenRouter, the same inference costs USD 20 and the card fee adds USD 1.10. OpenRouter can route the same model to different providers or service tiers, and fallback can change the endpoint that serves a request. The table therefore compares the displayed matching inference rate; it does not guarantee the bill from every default routing configuration.

Why Capriole uses a different unit

Capriole AI measures API usage in charged tokens. Each request draws from the same balance using this formula.
Uncached input and output count at 100%. Cached input counts at 10%. The example has no cached input, so 4 million input tokens plus 1 million output tokens consume 5 million charged tokens. Capriole deducts charged tokens from your balance. The table shows the price of a new personal Premium subscription alongside the direct API and OpenRouter charges for the example. Capriole AI Charged Tokens Explained provides three additional examples, rounding behavior, and the Premium quota boundary.

What USD 9 includes beyond the API balance

Capriole AI Premium includes the web workspace and 5M monthly API tokens for USD 9. Team members use their shared API balance. The USD 9 Premium plan includes:
  • Unlimited browser chat across the supported Premium model set
  • The first 5 million monthly API charged tokens
  • Access to the public API and maintained coding-agent integrations
  • One browser workspace with model switching, files, web search, and chat-storage controls
Before subscribing, signed-out visitors and Free users can try GPT-6.2 Astra, Claude Fable 5.1, Claude Opus 5.5, and Gemini 4 Carbon in free browser chat. That free flagship trial is separate from the API cost table above. If you want frontier-model chat and API access together, USD 9 covers the browser workspace and 5M API tokens a month. The same account also works with 10 documented coding-agent integrations. An active personal Premium member who needs another 5 million charged tokens can purchase one USD 8 top-up pack.

What changes the cost of a request?

The example excludes several pricing paths that can materially change cost:
  • Caching: Providers price cache writes and reads separately. Capriole counts cached input at 10% in its charged-token formula.
  • Long prompts: OpenAI charges higher rates for GPT-6.2 Astra, GPT-6 Astra, and GPT-6.2 Sol when a request has more than 272,000 input tokens, and Google does for Gemini 3.1 Pro Preview when a prompt exceeds 200,000 tokens.
  • Service tiers and batch processing: Free, Flex, Priority, Batch, or other provider tiers can differ from the paid Standard rates used here.
  • OpenRouter routing: The endpoint or fallback that serves a request can have a different price from the matching rate used in the table.
  • Tools and search: Server-side tools, grounding, and search can add request or usage charges.
  • Output-heavy workloads: The table uses an 80/20 input-to-output split. More output raises direct and OpenRouter costs faster for models with a large output-price premium.
  • Enterprise terms: Volume discounts, private offers, data residency, and negotiated commitments can replace public list prices.
Recalculate with your expected model, token mix, and features before choosing a provider.

The practical choice

The USD 9 subscription includes unlimited web chat, a browser workspace, 5M monthly API tokens, and coding-agent integrations. You can try Capriole AI in free browser chat or upgrade to USD 9 Premium for unlimited chat and an API key. For a wider product comparison, read Capriole AI vs OpenRouter. For official per-token prices of the current flagship models, see Claude API pricing, OpenAI API pricing, and Gemini API pricing. Developers can continue with the API quickstart. Capriole AI is an independent workspace, not affiliated with OpenRouter.

Official sources

Capriole plan, top-up, and charged-token facts on this page are maintained as first-party documentation. The comparison uses only public customer pricing and quota rules.
Last modified on November 18, 2026