Grok 4.7 API Access: Per-Token Pricing Without an X Subscription (2026)
LivePair AI ·
Grok 4.7 — xAI's frontier line — is in LivePair's text catalog alongside Grok 4.20. Both are callable through the standard chat-completions shape with a prepaid `lp_` key: no X subscription, no provider account, per-token billing against the credits you load.
Where Grok fits
Grok's strength profile leans toward conversational fluency, real-time-aware answers, and an unfiltered register that other frontier models sand down. For chat products and social-adjacent agents, that register is the feature. For structured extraction or code-heavy work, the same key reaches models tuned for that — Claude, GPT-6, DeepSeek.
Catalog ids: `x-ai/grok-4.7` and `x-ai/grok-4.20`. The provider-prefix convention means a model switch is a string change in an existing request body.
The call
Standard shape: `POST https://livepairai.com/v1/chat/completions`, `Authorization: Bearer lp_…`, body with `model`, `messages`, `max_tokens`. Any client or framework that accepts a custom `base_url` works — point it at `https://livepairai.com/v1`.
Set `max_tokens` deliberately — it doubles as the per-call spend ceiling, since billing quotes against that cap. `GET /v1/agent/models` returns the live catalog with in/out rates for every model, free, no key.
| Field | Value |
|---|---|
| Endpoint | POST https://livepairai.com/v1/chat/completions |
| Auth | Authorization: Bearer lp_… |
| Model ids | x-ai/grok-4.7 · x-ai/grok-4.20 |
| Spend cap | max_tokens = billed ceiling |
| Agent rail | x402 — USDC per call on Base, no key |
No subscription math
The usual Grok access path is a subscription tied to a consumer product. The API path on LivePair is different arithmetic: load $5+ of credits, pay per token, stop paying when you stop calling. For bursty or experimental traffic — prototypes, weekend builds, agent experiments — prepaid beats a monthly commitment by definition.
The same balance funds every model in the catalog, plus image and video generation on the media endpoints — one pool, no per-provider accounts.
Production routing pattern
The aggregator play that scales: tier your calls. Cheap lanes (`openai/gpt-5.4-nano`, `deepseek/deepseek-v4-flash`) handle classification and short turns; `x-ai/grok-4.7` handles the conversational surface where its register pays off; `anthropic/claude-sonnet-5.5` covers deep reasoning. One integration, three tiers.
The catalog endpoint is the routing table — fetch it, let users or heuristics pick. When a new model lands in the catalog, your picker sees it the same day.
FAQ
- How do I access the Grok API without X Premium?
- Through LivePair's aggregated endpoint: POST https://livepairai.com/v1/chat/completions with an lp_ key and model 'x-ai/grok-4.7'. Billing is per token from prepaid credits — no subscription involved.
- Is Grok 4.7 the same as Grok 4.20?
- They're sibling model lines in the catalog — 'x-ai/grok-4.7' and 'x-ai/grok-4.20'. Same key and request shape; pick by latency/quality preference from the live model list.
- Can the same key call Claude and GPT too?
- Yes — one lp_ key reaches 50+ text models: Claude Sonnet/Opus, GPT-6 Astra, Gemini 3.8, DeepSeek V4, Kimi K3, plus ~20 image/video generation models.