Sentiment Analysis API in India
Measure customer emotion in reviews, surveys, and feedback. Real ₹ pricing, batch API, no per-message SaaS limits.
Analyze Sentiment at Scale
Measure sentiment (positive, negative, neutral) in customer reviews, survey responses, social media feedback—all through one API endpoint, billed in rupees.
Why API-based sentiment beats UI tools
Volume: Analyze millions of texts in parallel; no monthly message caps. Transparency: Audit your prompts, logs, and model choices; full reproducibility. Integration: Pipe results into your data warehouse, CRM, or BI dashboard directly. Cost-friendly: Pay-as-you-go token pricing, no subscriptions.
Sentiment models ranked
| Model | Input/Output (₹) | Latency | Nuance |
|---|---|---|---|
| Qwen3 1.7B | ₹0 / ₹0 | <50ms | Basic |
| Qwen3.8-27B | ₹16.32 / ₹48.96 | 100–300ms | Good |
| Claude Sonnet | ₹201.6 / ₹1008 | 200–500ms | Excellent |
| Claude Opus | ₹504 / ₹2520 | 300–800ms | Premium |
Detect sentiment in 3 steps
Step 1: Prepare your prompt
Analyze the sentiment of the following text.
Respond ONLY with one of: POSITIVE, NEGATIVE, NEUTRAL, MIXED
Step 2: Batch-call the API
POST https://api.unoblox.ai/v1/messages
Authorization: Bearer ub-gw-...
{
"model": "qwen/qwen3-8-27b",
"messages": [
{"role": "user", "content": "Analyze sentiment:\n\n[Your review text here]"}
],
"max_tokens": 20
}
Step 3: Aggregate results — Count sentiment distribution, track trends over time, alert on spikes in negative feedback.
Real scenario: Analyze 100k reviews/month
- Extract reviews from your database.
- Batch 1,000 reviews per API call (10-20 second roundtrip).
- Collect results into a Postgres table.
- Build dashboards on sentiment trends by product, month, region.
- Cost: ~8M tokens total (≈70 input + ≈10 output tokens per review) → about ₹163 on Qwen3.8-27B, or about ₹2,419 on Claude Sonnet.
Advanced use cases
Multi-aspect sentiment: Extract sentiment for [product quality, pricing, delivery, support] separately.
Analyze sentiment for these aspects: Quality, Price, Delivery.
Respond as JSON: {"quality": "POSITIVE", "price": "NEGATIVE", "delivery": "NEUTRAL"}
Reason extraction: "Why is this negative?" Ask the model to explain in 1–2 sentences.
Sentiment: NEGATIVE
Reason: "Shipping took 30 days."
Emotion classification: Map sentiment + tone into emotions (angry, disappointed, happy, confused).
Frequently asked questions
Q: How accurate is sentiment analysis on mixed reviews? A: Smaller free models handle clearly positive or negative text well but struggle more on genuinely mixed or sarcastic reviews; Claude Sonnet is noticeably better on that harder tail. Test both on 100 hand-labeled reviews from your own data before picking one, and keep a MIXED label for genuinely ambiguous inputs.
Q: Can I analyze sentiment in Hindi or Hinglish? A: Qwen3 models handle Hindi well. Claude Sonnet also supports Hindi. Test on a sample of your data; results may vary by dialect.
Q: How do I handle sarcasm or context-heavy reviews? A: Sarcasm is hard for all models. Provide more context (product type, review length); use Claude Sonnet for best results. Fallback: hand-review top 10% of results.
Q: Can I combine sentiment with aspect extraction? A: Yes. In one call, ask the model to extract aspects (e.g., "shipping, quality, price") + sentiment for each aspect. Example JSON above.
Q: What's the latency if I need real-time sentiment (live chat)? A: Qwen3 1.7B: <50ms. Claude Sonnet: 200–500ms. For real-time, use the faster model; upgrade to Sonnet for accuracy if acceptable.
Q: How do I alert on urgent negative feedback? A: Flag NEGATIVE sentiment with keywords ("urgent", "refund", "error", "broken"). Route to a human queue for same-day review. Log for trend analysis.
Get started in rupees → https://unoblox.ai/sign-in
Start building in rupees
Call every major model through one OpenAI-compatible endpoint, billed in ₹ on a GST invoice.