Classer
high-performance classification API
Last verified:
What is Classer?
Classer is a high-performance AI classification engine designed specifically for turning messy data into accurate labels at scale. Unlike general-purpose LLMs built for human-like conversation, Classer strips away the
Classer pricing
Pricing model: Freemium
Pay for input tokens only with three priority tiers: Fast at $0.60 per 1M tokens with <200ms P95 latency for real-time UX and chatbots, Standard at $0.20 per 1M tokens with <1s P95 latency for background processing, and Fast Batch at $0.08 per 1M tokens with <15min P95 latency for bulk analysis and exports. First 10M tokens are free for new users. Save up to 8x vs public LLMs. Generative LLMs available with OpenAI-compatible inference: DeepSeek-V4-Flash at $0.10 input/$0.28 output per 1M tokens, Qwen3.6-35B-A3B at $0.15 input/$1.00 output per 1M tokens, and Kimi-K2.6 at $0.70 input/$4.00 output per 1M tokens.
Classer pros
- Beats GPT-5.4-mini accuracy on classification benchmarks
- Up to 100x cheaper than general-purpose LLMs
- Real-time latency under 200ms P95
- No prompt engineering required
- No training data needed - works zero-shot
- Zero schema violations - 100% valid outputs
- Zero hallucinations by design
- 60-second zero-shot setup time
- Auto-calibration feature that improves automatically
- Handles up to 50 million rows per batch job
- Non-technical Correct loop for improvement
- OpenAI-compatible inference for generative LLMs
- Pay for input tokens only - no reasoning token costs
- Custom model lives in your account after auto-improve
- First 10M tokens free for new users
- Save up to 8x vs public LLMs with priority tiers
- Works on 33 public datasets with proven benchmarks
- Deterministic latency unlike variable LLM response times
Classer cons
- Specialized only for classification - not for chat/conversation
- No free tier beyond the initial 10M tokens
- Fast tier at $0.60 per 1M tokens may be expensive for very high volume
- Batch processing has up to 15 minute latency
- Requires providing your own labels - not auto-discovering categories
- Limited to classification tasks - cannot generate open-ended text
- Custom model creation requires enabling auto-calibration manually
- Generative LLM options have higher costs for output tokens
Frequently asked questions about Classer
What is Classer and what does it do?
Classer is a high-performance AI classification engine that turns messy data into accurate labels at scale. It beats GPT-5.4-mini accuracy on classification benchmarks while being up to 100x cheaper with real-time latency. Unlike general-purpose LLMs built for conversation, Classerstrips away the chat interface to focus purely on classification tasks like sorting support tickets, detecting spam, routing phone calls, and tagging product images.
How does Classer compare to GPT-5.4-mini in accuracy?
Classer beats GPT-5.4-mini on top classification benchmarks with an average of 75.8% accuracy compared to GPT-5.4-mini's 67.7%, tested on 33 public datasets with zero training data. On specific datasets like LexGLUE ECtHR, Classer achieves 63.0% vs 15.6%, and on SMS Spam it achieves 97.0% vs 90.5%.
What are the pricing tiers and which should I choose?
Classer offers three priority tiers: Fast at $0.60 per 1M tokens with <200ms latency for real-time UX and chatbots, Standard at $0.20 per 1M tokens with <1s latency for background processing, and Fast Batch at $0.08 per 1M tokens with <15min latency for bulk analysis and exports. Choose Fast for instant answers, Standard for cost-effective background work, and Batch for the cheapest option when you don't need instant results.
Can Classer handle large-scale batch processing?
Yes. Fast Batch Processing can handle millions of results in 15 minutes with no concurrency scripts, retry loops, rate limit workarounds, or 24-hour waits. You can upload a file with up to 50 million rows per job and get labeled results back at whatever scale your pipeline needs. Batch is the cheapest way to run Classer if you don't need instant answers.
How do I improve Classer's accuracy over time?
Classer has a four-step improvement process: Start with zero-shot by passing your labels, monitor predictions in your console with confidence scores to spot edge cases, correct by adding class descriptions or labeling a few examples (or let a high-reasoning LLM do it automatically), and enable auto-improve which distills your data into a custom model that lives in your account. This happens in the background while you focus on your product.
Does Classer have schema violations or hallucinations?
No. Classer has zero schema violations and zero hallucinations. Pure classification means you get the right format every single time with 100% valid outputs, unlike generalist LLMs that have 15% schema violations.
Can I use Classer with OpenAI SDK?
Yes. Classer provides OpenAI-compatible inference for generative LLMs. You can point any OpenAI SDK at api.classer.ai/v1 and pick the model that fits, including deepseek-ai/DeepSeek-V4-Flash, Qwen/Qwen3.6-35B-A3B, and moonshotai/Kimi-K2.6 with various capabilities like 1M context, text+vision, and reasoning.
What AI tasks is Classer best suited for?
Classer is built for the 90% of AI tasks that don't need a chat interface. It's perfect for sorting support tickets, detecting spam, routing phone calls, tagging product images, classifying app reviews, sarcasm detection, rumour evaluation, question classification (TREC Question at 94.5%), SMS spam detection (97.0%), financial phrase classification (69.8%), and legal text classification like SCOTUS rulings (70.0%).
How do I get started with Classer?
You can get your API key in 30 seconds and start saving immediately. The first 10M tokens are free for new users. Start with zero-shot setup by just passing your labels, and the system works out of the box. No ML engineers needed - the non-technical Correct loop lets you improve the model without specialized expertise.