Is llmrelay cheaper than the Anthropic API?

Yes. Anthropic lists Claude Opus 4.7 at $5.00 per million input tokens and $25.00 per million output tokens. llmrelay sells the same model at $2.50 and $12.50 — exactly 50% of official list.

Does Anthropic have a cheaper option than list price?

The Batch API is 50% off list, which matches llmrelay pricing, but batch requests are asynchronous and unsuitable for interactive or agentic workloads. Prompt caching also lowers cost on repeated prefixes and is available on both.

// first-party vs relay

llmrelay vs Anthropic direct

Same model IDs, same API shape, half the token bill. The trade is a billing relationship, not model quality.

The price, per million tokens

Model

Anthropic in / out

llmrelay in / out

You save

claude-opus-4-8

$5.00 / $25.00

$2.50 / $12.50

50%

claude-opus-4-7

$5.00 / $25.00

$2.50 / $12.50

50%

claude-sonnet-5

$2.00 / $10.00

$1.00 / $5.00

50%

claude-fable-5

$10.00 / $50.00

$5.00 / $25.00

50%

Anthropic figures read from platform.claude.com/docs pricing, 26 July 2026. The $5/$25 Opus tier now covers 4.5 through Opus 5; only the deprecated Opus 4.1 and retired Opus 4 sit at $15/$75.

Where Anthropic direct is the right call

— You need an enterprise SLA, a signed DPA, or first-party support escalation.
— Procurement requires a direct vendor relationship for accounting or audit.
— You want day-zero access to brand-new models before any relay carries them.
— You need first-party-only features. Fast mode ($10/$50 on Opus 5 and 4.8) is Anthropic-only, and it errors on Opus 4.7 entirely.

Where llmrelay wins

If you're an indie developer, a small team, or an agency running agent loops, the API bill is the constraint and the billing relationship isn't. Halving the bill is the whole product.

You also get Claude, GPT, and Gemini families behind one prepaid balance instead of three separate accounts, and a published fork with a hidden-prompt neutralizer patch so you can see exactly what the relay does to your requests.

The honest caveat: Batch API

Anthropic's Batch API is also 50% off list, so on batch-friendly work Anthropic matches our price directly. Batch is asynchronous with a results window, which rules it out for chat, IDE assistants, and agent loops. If your workload is genuinely offline and non-interactive, Anthropic Batch is a fair alternative and we'd rather say so than pretend otherwise.

One thing that trips up cost estimates

Anthropic notes that Claude 4.7 and later use a newer tokenizer producing roughly 30% more tokens for the same text. The price per million didn't change, but your token count for identical input did. Any "cost per 1,000 documents" spreadsheet built on pre-4.7 token counts will understate spend on both platforms by about a third. The 50% gap between us holds regardless.

Migration cost: two env vars

# Before (Anthropic direct)
ANTHROPIC_BASE_URL=https://api.anthropic.com
ANTHROPIC_API_KEY=sk-ant-...

# After (llmrelay)
ANTHROPIC_BASE_URL=https://api.llmrelay.dev
ANTHROPIC_API_KEY=sk-llmrelay-...

Model IDs stay the same. The OpenAI-compatible endpoint works too if your SDK prefers it.

Same Opus, half the invoice

Point your existing client at llmrelay and watch the per-call cost drop by half. New accounts start with free credit, so you can verify before you top up.

Get API key →