Anthropic-compatible /v1/messages (₹ billing)

Claude SDK + unoblox: ₹-native billing for Claude (and Qwen, DeepSeek)

unoblox supports the Anthropic-compatible /v1/messages endpoint, so any code written against the Claude SDK works unchanged. Point it at unoblox, optionally swap the model to a cheaper open-weight option, and get monthly ₹ invoicing with GST input credit claimable — no foreign card, no separate billing relationship to manage.

Python: 2-line setup

from anthropic import Anthropic

client = Anthropic(
    base_url="https://api.unoblox.ai/v1",          # <- change #1
    api_key="ub-gw-YOUR_API_KEY"                   # <- change #2
)

response = client.messages.create(
    model="anthropic/claude-opus",  # or swap to qwen/qwen3-235b-a22b
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Explain GST in India"}
    ]
)

print(response.content[0].text)

That's it. No refactoring needed — the request shape, streaming events, and tool-use format all match what the Anthropic SDK already expects.

Feature parity with native Anthropic

Feature	unoblox	Anthropic
/v1/messages	Yes	Yes
Streaming	Yes	Yes
Tool use	Yes	Yes
Vision	Yes (Claude + Qwen)	Yes
Batch API	Coming soon	Yes
Prompt caching	Coming soon	Yes

Model choices (all support /v1/messages)

Keep Claude for the calls where its judgment matters, and route everything else to a cheaper model through the same endpoint and message format:

• anthropic/claude-opus ₹504 / ₹2520 per 1M tokens (quality)
• anthropic/claude-sonnet ₹201.6 / ₹1008 per 1M tokens (balance)
• qwen/qwen3-235b-a22b ₹9.07 / ₹55.44 per 1M tokens (cheap, vision)
• deepseek-ai/deepseek-v3.2 ₹26.21 / ₹38.30 per 1M tokens (reasoning)

Migrating a live integration

1. Get a key. Sign in and create an ub-gw-… key — no separate Anthropic account needed alongside it.
2. Swap the base URL and key in your existing Anthropic client config; leave the rest of your code untouched.
3. Test one real call against your current model (anthropic/claude-opus or anthropic/claude-sonnet) to confirm parity, then decide per-endpoint whether to keep Claude or route to a cheaper model.

Why swap to an open model?

• Cost. Qwen and DeepSeek cost a fraction of Claude Opus per token, so high-volume calls (classification, first-pass drafts, bulk summarization) get dramatically cheaper without a code change.
• Sovereignty. Open-weight models run without a dependency on a single foreign SaaS vendor, which matters for regulated or security-conscious teams.
• Compliance. One monthly ₹ GST invoice from an Indian entity, with input tax credit claimable — easier for finance to reconcile than a foreign card statement.
• Flexibility. Pick the model per request without touching your client setup — keep Claude for judgment calls, route bulk work elsewhere.

Frequently asked questions

Q: If I use Qwen instead of Claude, will my code break? No. Both support the same Anthropic message format. Swap the model ID, test once, deploy.

Q: Can I use tool use with Qwen via this API? Yes. The Anthropic SDK's tool-calling handling works with any /v1/messages-compatible model on unoblox.

Q: Is there vendor lock-in? No. Any /v1/messages endpoint works the same way. You can point back at Anthropic's own API in one line if you ever need to.

Q: Do I get billing details like Anthropic provides? Yes. Token counts and cost per call appear in the unoblox dashboard (/dashboard), and you can download invoices anytime for accounting.

Q: Can I mix Claude and Qwen in the same app? Yes. Use two clients with different base_url/model pairs, or one client and switch the model string per call.

Q: What about streaming and vision together? Both work together. Qwen vision plus streaming works natively, and Claude models support the same combination on unoblox.

Get started in rupees → https://unoblox.ai/sign-in