Skip to content

Models Catalog

Welcome to our Models Catalog, where you can explore our collection of cutting-edge compressed language models. CompactifAI API delivers a new class of compressed, ultra-efficient AI models engineered for performance, sustainability, and adaptability. These dramatically reduce compute and energy costs while maintaining enterprise-grade accuracy and speed. It adapts seamlessly to diverse environments, ensuring consistent, high-performance AI. With turnkey software and an intuitive API, it offers cost effective inference enabling enterprises to innovate faster, scale cost-effectively, and unlock competitive edge.

  • Dramatic Cost Reduction: Up to 70% lower inference costs compared to uncompressed models through optimized resource utilization
  • Massive Throughput Gains: Process up to 4x more requests per second with compressed models requiring fewer computational resources
  • Low-Latency Inference: Achieve faster response times due to reduced model size and optimized memory usage
  • Minimal Quality Loss: Advanced compression techniques preserve model performance with typically <5% benchmark difference
  • Superior Concurrency: Support significantly more simultaneous users and requests with the same hardware resources
  • Resource Efficiency: Reduced memory footprint and computational requirements enable better hardware utilization

Use the following API model identifiers in your requests:

Model Name Original Architecture Model ID Available
GLM 5.2 GLM glm-5-2 Yes
GLM 5.3 GLM glm-5-3 Yes
Quasar 1.2 358B GLM quasar-1-2-358b Yes
Carina 60B GPT-OSS carina-60b Yes
Qwen 3.8 27B - qwen-3-8-27b Yes
Whisper Large V3 Turbo Slim by CompactifAI Whisper Large v3 Turbo cai-whisper-large-v3-turbo-slim Yes

The models are served from several regions, each with its own endpoint:

Region Base URL
EU https://api-eu.compactif.ai/v1
US 1 https://api-us.compactif.ai/v1
US 2 https://api-usw2.compactif.ai/v1

The context length is the maximum number of tokens a request can use in total, your prompt plus the tokens the model generates, and it is set per model and per region. A dash (-) means the model is not served from that region. Values are rounded down to the nearest round figure (k = thousand, M = million tokens).

Model EU US 1 US 2
glm-5-3 500k - 1M
glm-5-2 - 200k 200k
quasar-1-2-358b 200k - 1M
carina-60b 100k 100k -
qwen-3-8-27b 1M 1M -

A request whose prompt plus requested output exceeds the limit is rejected with a 400 error and the code context_length_exceeded.