Docs · Routes and pricing
Routes and the auto route
One route per provider: what each API-key route sells, its price next to the provider list price, and how one Auto-route key covers every model.
Every API key belongs to one route, the Group field when you create a key (Edit changes it later). The route decides which models the key can call and which fixed USD price applies. Every model is sold on exactly one route, so the price of a model is the same whichever key calls it.
The routes
Claude, OpenAI and Grok each have one route at 80% of the provider's list price, in every lane: input, output, cache read and cache write. Every price table on the site shows the provider's list price next to the TopxAI price.
- Claude (
claude-shared):claude-opus-5-5,claude-sonnet-5-5andclaude-fable-5-2. - OpenAI (
openai-shared):gpt-6.2-astra,gpt-6.2-solandgpt-6.1-terra. - Grok (
grok-shared):grok-4.7. - Grok video (
grok-official):grok-imagine-video-1.5, billed per second.
The route ids are historical. Until 2026-11-18 Claude, OpenAI and Grok had a second route each; keys on those moved to the provider's route on their own, and requests they served keep their original price in the usage log.
The Claude models and GPT-6.1 Terra keep a backup line. When the main line fails before it answers, the request moves to the backup at the same price, so nothing changes for you. A stream that already started is not resent. GPT-6.2 Astra, GPT-6.2 Sol and Grok 4.7 run on one line for now.
For a limited time Claude also has a second route, Special offer (claude-special), for claude-opus-5-5 only: $0.45 input / $0.90 output / $0.10 cache read / $0.20 cache write ($0.30 for a 1-hour write) per million tokens. It is often unavailable for long stretches with no availability guarantee, adds its own system prompt to every request, has no fallback and is never chosen by Auto route. Its terms are on the Claude Opus 5.5 special offer page.
Four model families have one route at a fixed price. The GLM, Kimi and DeepSeek routes keep their historical -official ids, but every model on them is served from a private deployment, not the provider's own API, at prices that are the same at every hour (USD per million tokens: input / cache read / output):
- GLM:
z-officialcarries GLM-5.3-Abliterated, a fine-tuned GLM-5.3, text only, at $3.00 / $0.45 / $6.00, and GLM-5.3-Flash-Abliterated, with images, at $0.36 / $0.075 / $1.50. - Kimi:
kimi-officialcarries kimi-k3.1-pro-Abliterated, Kimi K3.1 Pro with the refusal alignment removed, at $3.25 / $0.325 / $6.50 (since 2026-11-19; it replaced kimi-k3.1-pro). - DeepSeek:
deepseek-officialcarries deepseek-flash-Abliterated at $0.36 / $0.075 / $1.35 (since 2026-11-19; $0.42 / $0.105 / $1.50 before) and deepseek-flash-fast, DeepSeek V4.1 Flash 1017 on standard weights, at $0.02 / $0.004 / $0.16. - Jev (TypeSafe's System One model):
typesafe-official, viaPOST /v1/systemone. Modeljev-1.14.0(aliasesjev-latest,jev-previewbill the same): $0.055 per million input tokens, output $0.
GLM-5.3-Abliterated, deepseek-flash-Abliterated and, since 2026-11-12, Jev also keep a costlier standby. A request moves to it only in the rare case that the main line fails before answering, and is then billed at the model's backup price, which the price tables list in a Backup deployment row; see how requests are billed.
Image models (gpt-image-2.6-sunburst, gpt-image-2.6-flare) sit on the image route, priced per image.
Prices differ per lane (input, output, cache read, cache write), and OpenAI and Grok add a long-context tier from 272,001 and 200,000 input tokens. Expand a route's model-count button in the Group list for its USD per 1M tokens table.
Two refusals happen before anything is forwarded or charged, both HTTP 403 permission_denied: a model outside the catalogue gets "Model is not offered by this service"; a model your route does not sell gets "Model is not available on this API key route".
Auto route
Pick Auto route and one key covers the catalogue. Each request is billed on its model's route at that route's price, exactly as with a key created for the route: Claude models on the Claude route, image models on image, and so on.
A model without a configured price gets the same 403 as above. If no channel can serve the model right now, the request fails with HTTP 503 upstream_unavailable, "The model provider is temporarily unavailable. Please retry." Nothing is charged either way.
The usage log shows the concrete route next to the key name; auto_route: true appears only in the API's log record, never auto as a route.
Related
- Create an API key and choose a route
- How requests are billed
- Every model's price
- TopxAI vs the providers' own APIs: price, retention and what stays direct
Previous: Hosted MCP servers: models, images and video in any MCP client · Next: How requests are billed
Other languages: 简体中文