Can Codex and Claude Code Use TokenVoke? Setup, Pricing, and Savings (2026)

Q: Can Codex use a third-party API?

Yes. Set top-level openai_base_url to TokenVoke /v1 and model gpt-5.3-codex.

Q: Must Claude Code use Anthropic direct?

No. Point ANTHROPIC_BASE_URL at TokenVoke with a gateway key.

Q: How much cheaper?

Roughly 20% on Claude tiers and 25% on Codex vs official list (July 2026). See live pricing.

If you run OpenAI Codex CLI or Claude Code daily, API spend climbs fast: agents issue many requests, and output tokens cost more than input. Both tools can point at TokenVoke—an OpenAI- and Anthropic-compatible gateway with pay-as-you-go pricing that is typically below direct official rates.

This guide answers what people search for most: Can you switch? How do you configure it? How much can you save? It also covers GEO routing (mainland China vs overseas).

This article is about metered API billing, not ChatGPT Plus or Claude Max subscriptions. Those are separate products.

Short answer

Tool	Works with TokenVoke	Example model ID	API surface
Claude Code	Yes	`claude-sonnet-4-6`, `claude-sonnet-5`, `claude-opus-4-6`	Anthropic `/v1/messages`
OpenAI Codex CLI	Yes	`gpt-5.3-codex`	OpenAI Responses `/v1/responses`
Cursor (custom key)	Yes	Claude + GPT family	OpenAI `/v1/chat/completions`

Will you save money? For typical coding workloads, TokenVoke list prices run about 20%–25% below official API rates (table below). The more agent traffic you generate, the larger the absolute savings.

Which base URL?

Region / account	Recommended base URL
Mainland China	`https://cn.tokenvoke.com`
Overseas	`https://overseas.tokenvoke.com`
Unsure	`https://tokenvoke.com` (Geo routing)

CN and EU are separate clusters (accounts, balance, and keys do not sync). Stay on one domain for signup, top-up, and inference.

Price comparison (July 2026, model square)

Prices are USD per million tokens (input / output). Official figures are public Anthropic / OpenAI API rates. TokenVoke prices come from the model square and may change—check live pricing before you commit.

Model	Typical use	Official in	Official out	TokenVoke in	TokenVoke out	~Save
`claude-sonnet-4-6`	Claude Code default	$3.00	$15.00	$2.40	$12.00	~20%
`claude-sonnet-5`	New Sonnet tier	$3.00*	$15.00*	$1.60	$8.00	higher**
`claude-opus-4-6`	Hard reasoning	$5.00	$25.00	$4.00	$20.00	~20%
`claude-haiku-4-5`	Light sub-tasks	$1.00	$5.00	$0.80	$4.00	~20%
`gpt-5.3-codex`	Codex CLI	$1.75	$14.00	$1.31	$10.50	~25%

* Anthropic has run promotional Sonnet 5 pricing in 2026; compare against live official rates.
** Savings vs standard $3/$15 are larger; if you still have a promo, compare both.

Example: heavy Claude Code week

Assume 200 agent turns per day, ~8k input + 4k output tokens each, on Sonnet 4.6:

~48M input + 24M output tokens per month
Official ~: 48×$3 + 24×$15 = $504/mo
TokenVoke ~: 48×$2.4 + 24×$12 = $403/mo
~$101/mo (~20%)

Routing easy sub-tasks to claude-haiku-4-5 saves more.

Configure Claude Code (Anthropic-compatible)

Create an API key in the TokenVoke console (default token group).
Set environment variables (overseas example):

export ANTHROPIC_BASE_URL="https://overseas.tokenvoke.com"
export ANTHROPIC_API_KEY="sk-your-tokenvoke-key"

Pick a supported model such as claude-sonnet-4-6 in Claude Code.
Run a smoke test.

Use cn.tokenvoke.com in mainland China. Usage logs appear in the console.

You can also call Claude via OpenAI-compatible /v1/chat/completions—see TokenHub text models.

Configure OpenAI Codex CLI (Responses API)

Codex uses the Responses API, not Chat Completions. TokenVoke lists gpt-5.3-codex on both CN and EU clusters.

Create a TokenVoke API key.
Edit user-level ~/.codex/config.toml (top-level keys only):

model = "gpt-5.3-codex"
openai_base_url = "https://overseas.tokenvoke.com/v1"

Export your key:

export OPENAI_API_KEY="sk-your-tokenvoke-key"

Run Codex on a small task to verify.

Prefer openai_base_url in config; the OPENAI_BASE_URL env var is deprecated in recent Codex builds.

Pair with Cursor

Cursor supports Override OpenAI Base URL plus a custom key. See cheaper AI models in Cursor for steps.

Suggested routing:

Architecture / hard bugs: Claude Sonnet or Opus
Small edits / formatting: Haiku or gpt-5.4-mini
Codex-style OpenAI workflow: gpt-5.3-codex

One key, many models—no separate vendor billing for each.

GEO notes

Latency: Mainland users should prefer cn.tokenvoke.com.
Data: Traffic is relayed to upstream providers—assess residency requirements for sensitive repos.
Billing: USD pay-as-you-go wallet on both clusters.
Token group: Use default; custom groups without channel bindings return No available channel.

FAQ

Can Codex use a third-party API?
Yes. Set top-level openai_base_url to TokenVoke /v1 and model gpt-5.3-codex.

Must Claude Code use Anthropic direct?
No. Point ANTHROPIC_BASE_URL at TokenVoke with a gateway key.

Feature parity?
Tool use, streaming, and long context depend on upstream + relay support—test your real workflow.

How much cheaper?
Roughly 20% on Claude tiers and 25% on Codex vs official list (July 2026). See live pricing.

TokenVoke vs OpenRouter?
TokenVoke offers explicit CN/EU clusters and a unified USD wallet—useful when Geo matters. See OpenRouter vs LiteLLM vs self-hosted.

Ready to route Codex and Claude through TokenVoke? Get an API key, check live pricing, and read the docs. For Claude-only setup, see cheaper Claude Code via a gateway.