Switch AI provider without a rewrite

Most AI APIs lock you in. Change from GPT to Claude? Rewrite your prompts, adjust token limits, test everything. Not with unoblox. It's OpenAI-compatible, so switching models is one config change.

The Problem: Vendor Lock-In

• OpenAI: Its own API contract, no GST invoice, an international card required, rate limits that vary by tier
• Anthropic: A different message format, a separate account, its own token counting, no rupee invoicing
• Together AI: Narrower model selection, US-based routing, the same international-payment friction
• Azure OpenAI: Enterprise-only onboarding, more setup steps, data-residency terms you have to dig for

Each requires app rewrites. unoblox fixes this.

How OpenAI Compatibility Works

unoblox wraps every model—GPT, Claude, Qwen, DeepSeek, Llama, Gemma, Kimi—behind a single OpenAI SDK interface. Swap models by changing one line:

// Switch from GPT to DeepSeek without rewriting
client = OpenAI(api_key="ub-gw-...", base_url="https://api.unoblox.ai/v1")
response = client.chat.completions.create(
  model="openai/gpt-4o",  // Change this one line
  messages=[...]
)

Same parameters, same streaming, same tools—all supported across providers.

Price Comparison (same endpoint)

Model	₹ per 1M input tokens	Use case
Qwen3.8-27B	FREE	Agents, retrieval, low-stakes
DeepSeek V4 Flash	₹9.11	Budget coding, translation, summary
Qwen3 235B	₹9.11	Reasoning, RAG (text-only)
GPT-4o	₹253.07	Complex tasks, broad knowledge
Claude Opus	₹506.14	Long-context, nuanced reasoning

Test cheaper models in dev; scale to premium in production—all without vendor rewriting.

Frequently asked questions

Do I lose Claude/GPT features (tools, JSON mode)?

Mostly no, but support varies by model. Streaming, tool use and JSON mode are widely supported; image input is only available on models whose page lists it. Check each model's page on /models.

Will switching models break my app?

Only if your model uses genuinely unique behavior (e.g., Claude's thinking parameter). unoblox normalizes core OpenAI semantics, so standard prompts port seamlessly.

How do I test a new model safely?

Use A/B routing: send 5% of requests to the new model in a feature flag. Monitor latency + cost. Full rollout takes hours, not weeks.

Can I use multiple models in production simultaneously?

Yes. Route different queries to different models via unoblox (e.g., simple queries → Qwen3.8-27B, complex → GPT-4o). Load-balanced by your code.

What happens if unoblox is down?

Implement local fallback + retry logic. unoblox targets 99.95% uptime; when down, your code can timeout and alert your team.

Do tool definitions need per-model adjustment?

Rarely. OpenAI tool schema works across Claude, GPT, DeepSeek, Qwen. Test once, deploy everywhere via unoblox.

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