Which Fikra API model is right for my workload?
Fikra API abstracts the complexity of hardware provisioning by routing your requests across highly optimized open-weights models and proprietary endpoints seamlessly. The registry allows you to explicitly declare your required pipeline based on latency, context window, and reasoning depth.
Model Family Overview
Pricing scales with compute intensity. Fikra uses a Base rate of $1.00 per 2M tokens ($0.50/M) and applies multipliers for advanced pipelines. To select a model, inject the exact String ID into the model parameter of your JSON request body.
| Model String ID | Context Window | Output/Reasoning | Target Use Case & Tier |
|---|---|---|---|
| fikra-flash | 8,192 tokens | 4,096 max output | Default. Ultra-low latency, real-time chat. Base Tier ($0.50/1M) |
| fikra-fast-8b | 8,192 tokens | 4,096 max output | High-volume automated classification and routing. Base Tier ($0.50/1M) |
| fikra-pro-20b | 32,768 tokens | 8,192 max output | General production, RAG pipelines, structured JSON. General Tier ($2.00/1M) |
| fikra-pro-120b | 128,000 tokens | 16,384 max output | Large document synthesis, logical deduction. Heavy Synthesis ($3.00/1M) |
| fikra-qwen-max | 1,000,000 tokens | 128k output / 256k CoT | Thinking Engine. Massive data ingestion & deep reasoning. Reasoning Tier ($5.00/1M) |
| fikra-coder Upcoming | 128,000 tokens | 16,384 max output | Full-repository code analysis and AST validation. Coding Engine ($3.50/1M) |
| fikra-auto | 32,768 tokens | 8,192 max output | Auto-redacting standard production pipeline. General Tier ($2.00/1M) |
| fikra-auto-pro | 1,000,000 tokens | 128k output / 256k CoT | Auto-redacting enterprise pipeline with massive context. Reasoning Tier ($5.00/1M) |
Deep Reasoning Advisory: Qwen Max
The fikra-qwen-max model activates internal Chain of Thought (CoT) processing. It is engineered for enterprise data ingestion with a massive 1,000,000 token input context window (991k raw input, 983k with thinking mode active).
Qwen Max performs up to 256k tokens of hidden CoT reasoning before delivering a final output stream. These reasoning tokens are billed identically to output tokens at the Reasoning Tier rate ($5/M). Developers must allocate ample max_tokens to prevent truncation mid-thought and extend HTTP client timeouts to account for the increased Time-To-First-Token (TTFT) delay.
Upstream Anonymity & Resilience
Fikra API automatically abstracts provider saturation. When a specific upstream node (e.g., Groq) becomes rate-limited or times out, the router automatically walks an internal fallback array (including OpenAI OSS pathways and Alibaba nodes) to keep your connection alive. Fikra scrubs all proprietary provider headers and system_fingerprint identifiers before returning the payload to you, ensuring zero leakage of backend architecture while maintaining 99.9% uptime.