Black label · Built for Claude Code
Black-label quality. Opus first.
OpusVIP is a premium Claude Opus relay built for Claude Code: a black-label, production-grade lane that real production projects run on every day. Set two environment variables and you're in. It's fast, every call is logged for you to check, and usage is billed at Anthropic's list prices.
This lane was built for production projects that run on Claude Opus every day. So it has to be fast, stable and accountable for every cent. OpusVIP opens it to you, unchanged.
The model you ask for is the model that answers. Opus 5.5 is the flagship workhorse; Fable 5.1, Sonnet 5.5 and Haiku 5.5 are one switch away, each with a clear job.
Streaming output and fast responses on a production-grade lane. Production workloads share it, so its speed is held to production standards.
You get the same resource pool and routing the production projects use. There is no second-tier route set aside for outside users.
How it works
Claude Code sends requests to api.opusvip.dev. OpusVIP handles auth, billing and logging, then forwards each request unchanged to the model you picked.
Claude Code
Your terminal or IDE
api.opusvip.dev
Claude Opus
Anthropic models
Sign up for the OpusVIP console. Google sign-in is supported.
On the console's token page, create a key that starts with sk-.
Point Claude Code at OpusVIP and work as usual.
export ANTHROPIC_BASE_URL=https://api.opusvip.dev
export ANTHROPIC_AUTH_TOKEN=sk-your-key
claude -p "hello"For Windows, or to make the setup permanent in settings.json, see the setup docs.
Models
Every model is billed at Anthropic's list price. Prices are in credits per 1M tokens (1 credit = $0.01 at list price), with what you pay and the USD list price underneath.
claude-opus-5-5 Flagship workhorse: complex coding, architecture, long tasks and hard bugs. 1M context.
credits / 1M tokens
claude-fable-5-1
Anthropic's most capable model: the hardest reasoning and long-horizon agent work
credits / 1M tokens · List $10 / $50
claude-sonnet-5-5
Balanced partner: subagents, bulk edits and everyday development
credits / 1M tokens · List $2 / $10
claude-haiku-5-5
Fast and light: small tasks, session titles and quick questions
credits / 1M tokens · List $0.10 / $0.50
Prompts over 100K use the long-context rate: input 50 · output 250
15% of list price, with usage metered at Anthropic's rates
Cache-write prices on the cards are the 5-minute rate; the full table also lists 1-hour caching. Failed requests are never billed. Top up in the console with Stripe (card, Alipay or WeChat Pay) from $1.50.
Billing and transparency
Billed at Anthropic's list prices, item for item, so you can check every line against the official price list. You pay 15% of list price.
Cache writes and reads are billed at Anthropic's cache rates, so the context Claude Code resends during a long session isn't charged at full price every turn.
Upstream errors and requests that don't complete are never charged.
The console records the model, input / output / cache tokens and credits charged for each request, so you can see at a glance whether caching is working.
Tuned for the official Claude Code CLI, so behavior stays predictable. Its default aliases map straight to the four models, and two environment variables are all it takes.
No monthly fee and no plan to commit to. You pay only for what you use.
Open a question to see the answer. More on the FAQ page.
claude-opus-5-5, the most capable claude-fable-5-1, and two partners, claude-sonnet-5-5 and claude-haiku-5-5. All four have a 1M context window and up to 128K output tokens (Haiku 5.5 prompts over 100K tokens are billed at Anthropic's long-context rate). Full prices are on Models and pricing.ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN, are set, Claude Code starts on Opus 5.5. Type /model opus, /model fable, /model sonnet or /model haiku in a session to switch, or pass the --model flag at startup. See Choose and switch models.Sign up for the console, create a key and top up with Stripe (card, Alipay or WeChat Pay) from $1.50.