Choose models and verify pricing and usage

Model availability, service groups, and prices can change independently. Read the model list with the intended API key, then compare the same model and service tier on the pricing page.

The pricing page calls each source and price choice a service tier. API keys and Usage call the same value a service group. This page uses the term shown in the step you are completing.

When only one service tier is available, the pricing page shows its prices directly without requiring a selection. This simpler display does not change an existing API key’s group or account permissions. Check the model scope and actual price for each existing key.

Benefit group migration deadline

The benefit group is closed to new selection. Existing benefit-group access remains available for now, but it will stop completely after September 30, 2026, according to the retirement notice. Migrate to another available group before the deadline. Keeping a key’s existing settings does not extend the group’s availability.

Before migrating, check the destination group’s models, prices and available quota. For Astra, Fast consumes 2.5 times the quota of Standard; choose the processing mode for your workload. After migrating, send a small request using your application’s protocol. Check the model, service group, cost and result in Usage before restoring normal traffic.

If your account has no other available group, or the target model is unavailable, contact support to confirm an available option.

Read available models

curl https://router-api.xiu.ai/v1/models \
  -H "Authorization: Bearer $XIUROUTER_API_KEY"

The response depends on:

• The service groups available to the account.
• The group selected by the API key, including automatic selection.
• The model scope applied to the API key.
• Whether a usable channel currently serves the model.

Two keys under the same account can return different model lists. Save exact model IDs in production configuration instead of matching only a provider or display name.

Antigravity resources and image tiers

Users do not select an internal channel. Native Gemini requests are routed to Antigravity by request path and model.

• Gemini 3.8 Flash can select low, medium, and high through the base model and thinkingLevel; use exact gemini-4-argon-low, gemini-4-argon-medium, or gemini-4-argon-high IDs when an Argon tier must be guaranteed because the base model can fall back to high through upstream mapping.
• gemini-3.1-flash-image has been verified at 1K, 2K, and 4K; 2K + 16:9 returned 2752×1536 and 4K + 1:1 returned 4096×4096.
• Image generation and editing use the image model’s current per-image price. Verify input images, output count, output size, and cost separately. The current max-group effective prices are $0.02 for 1K, $0.03 for 2K, and $0.06726 for 4K; use the pricing page and Usage for other groups.
• Family grouping on the pricing page changes presentation only; requests still use exact model IDs.

If a client cannot express thinkingLevel, use an exact -low, -medium, or -high ID. Otherwise prefer the base model and request parameter, then read back upstream_model.

Cache boundary

Channel 39 does not currently provide Gemini explicit context caching. POST /v1beta/cachedContents returns 404 and does not charge the key; Antigravity accounts do not satisfy Sub2API’s explicit-cache account contract. Do not treat repeated prompts or one successful request as a cache hit. Explicit caching requires a separate Gemini-platform account and a channel that supports cachedContents, with cache_read_tokens and cache charges verified in Usage.

Gemini model families

The pricing page groups Gemini reasoning levels into one model-family row so compatibility suffixes do not look like separate models. Choose the request shape according to the client:

• For Gemini 3.x, prefer the base model ID and send reasoning_effort.
• Argon low, medium, and high are native Antigravity tiers. Use an exact tier ID when the tier must be guaranteed because the base Argon name can fall back to high through upstream mapping.
• Grouped pricing display does not change the exact IDs returned by /v1/models or the API key’s model scope.

After changing the model or reasoning level, send a small request and check the final model, protocol, service group, and charge in Usage.

Check the protocol

The public pricing catalogue lists supported endpoint types. Verify the final path with the intended API key:

The gateway also has image, Embeddings, audio, Rerank, and Moderation routes. Availability depends on the model and service source. Files, Fine-tuning, and image Variations are not general XiuRouter capabilities.

Compare prices

Open Models and pricing, then:

1. Search for the exact model ID.
2. If the page offers multiple service tiers, select the one used by the API key. With one tier, read its prices directly.
3. Confirm that the key can access the model. A public price listing does not grant the account access to every model.
4. Check whether billing is token-based or per request.
5. For token billing, compare input, output, cache-read, and cache-write rates.
6. If the row has multiple context tiers, check the input-length condition and all token rates for each tier.
7. Use Fast or Flex only when the selected model row lists that processing mode.
8. Use the provider price for comparison only. XiuRouter bills the current service-tier price recorded in Usage.

A service tier selects a source and price. It does not promise a fixed speed, quality level, or SLA. An automatic group can retry another group after a failure; the group that completes the request determines the price.

Dynamic context pricing

Some models switch the complete token price after the input context crosses the condition shown on the pricing page. Input, output, cache-read, and cache-write rates can all change together.

Use the full input length to identify the tier. Do not choose a tier from non-cached input alone, and do not charge cached tokens again as ordinary input when the cache rate is listed separately.

Some Responses models may also list a service_tier processing mode:

These examples use priority for Fast requests. Send a processing mode only when the selected model and service tier list it. The multiplier applies to the matched token price. Hosted tools and other surcharges are recorded separately in Usage.

Verify Usage

Check costs for a date range

Sign in and open Usage to check existing calls. You do not need to send another request.

1. Open the time-range picker. Choose Today, Yesterday, Last 7 days, Last 30 days or This month to apply a preset.
2. For a custom period, select the start and end dates and times in the calendar, then select Apply.
3. Review the token totals and recorded costs for that period. To find specific calls, filter the request list by API key, model or outcome.

Dates use your browser’s local time. Last 7 days and Last 30 days include today; This month starts on the first day of the month. Refreshing a custom range keeps its start and end times unchanged.

The summary at the top covers the selected time range. API key, model and outcome filters affect the request list, not those summary totals.

If no records appear, check the date range and filters. A request without a recorded cost is not necessarily free; refresh later and inspect its details.

Check an individual request

After the request, open Usage in XiuRouter. Check:

• Request time and request ID.
• API key name and public ID.
• Model and final service group.
• Input and output tokens.
• The matched context tier and request rule for dynamic models.
• Hosted-tool calls and separate surcharges.
• Recorded cost and cost status.
• Streaming state.
• Time to first token, total duration, and request result.

Refresh a failed or unknown record before diagnosing it. Missing request IDs, token counts, or charges are unknown facts; do not replace them with zero or a client estimate.

Usage shows request statistics, not conversation content. XiuRouter does not store model conversation content. Usage statistics and model performance data are retained long-term, and model providers apply their own service data policies. See Data handling and privacy for the scope.

Recommended flow

1. Choose the model on the pricing page. If multiple service tiers are available, select the one used by the key.
2. Create a project-specific API key with a model scope and quota.
3. Confirm that GET /v1/models returns the target model for that key.
4. Send a small request through the protocol used by the application.
5. Match the model, group, tokens, charge, and result in Usage.
6. Increase concurrency, context length, and tool permissions only after the small request passes.

Sources and review date

The single-tier pricing display, processing-mode request values, data-policy link, and benefit group retirement notice were reviewed on September 13, 2026. The migration deadline and Astra Fast quota multiplier come from that notice. The model-list, API-key scope, and Usage guidance retains the previously verified API and field descriptions. Use the current console for model availability, groups, and prices.

Date-range instructions and the English notice link were reviewed on September 17, 2026. Range labels, date boundaries and refresh behavior were checked against the current product implementation and deployed resources.

Antigravity native Gemini image sizes, reasoning levels, tools, and per-image billing were added on November 7, 2026 and verified against production usage records.