Compute & Models
In Aitok, "Compute" refers to the processing power and model resources that drive AI inference, dialogue, and Agent tasks.
Whether you need peak reasoning intelligence or cost-effective high-volume throughput, Aitok's Compute Market provides a versatile catalog of world-class models.
Compute Market at a Glance
Navigate to Compute (/hub/compute) in the sidebar to inspect all available model services:
- Model ID: The exact string to pass in your API requests or tool configuration (e.g.,
gpt-4o,claude-3-5-sonnet-20241022,llama-3.3-70b). - Pricing: Clearly displayed pricing per million tokens (1M tokens) for both Input Tokens and Output Tokens.
- Context Window: The maximum token capacity supported in a single session (e.g., 128k, 200k tokens).
- Protocol & Format: Clearly displays the native protocol format supported by that specific model provider.
- Operational Status: Live health and latency metrics for each model endpoint.
Supported Model Families
Aitok aggregates leading foundation models under one unified umbrella:
1. Anthropic Claude Series
- Claude 3.5 Sonnet: The gold standard for programming, tool-use/function calling, and multi-step Agent reasoning. Highly acclaimed for precision and instruction adherence.
- Claude 3.5 Haiku: Blazing-fast inference combined with strong cost-efficiency, ideal for quick Q&A and text transformations.
2. OpenAI GPT Series
- GPT-4o: The flagship multimodal foundation model with text and vision intelligence across numerous domains.
- GPT-4o-mini: Fast, cost-efficient powerhouse for high-concurrency tasks and structured data parsing.
- o1 / o3-mini: Designed for complex reasoning and deep chain-of-thought problem solving in mathematics, science, and intricate logic puzzles.
3. Meta Llama Series
- Llama 3.3 70B: The leading open-weights architecture, matching proprietary frontier performance with exceptional efficiency across dialogue and reasoning.
- Llama 3.1 405B: The hyper-scale open model designed for high-capability reasoning and synthetic data generation.
4. Mistral AI & Google Gemini
- Mistral Large: Renowned for sophisticated reasoning, native multilingual versatility, and advanced coding benchmarks.
- Gemini 1.5 Pro: Features industry-leading multimillion-token context windows for complex multi-modal research.
Pricing Rules & Transparent Billing
- Pay-as-You-Go: No forced high monthly subscription fees. Every call is billed solely on the actual tokens sent (input) and received (output).
- Real-Time Settlement: Immediately following completion of a model call, token usage is calculated against the published rates and deducted from your balance in real time.
- Transparent Currency: Avoid unpredictable cross-border exchange rates or credit card processing fees; all prices are settled transparently in your primary balance.
Model Recommendations for Agent Workflows
When pairing models with AI Agents (coding assistants, research copilots, automation bots), match models to your primary needs:
| Task Type | Recommended Models | Key Advantage |
|---|---|---|
| Autonomous Coding / Agent Tools | claude-3-5-sonnet or gpt-4o | Superior function calling, dependable instruction following, low error rates |
| Deep Reasoning & Logic | o1 or o3-mini | Built-in step-by-step thinking for complex problem sets |
| Long Document Research | claude-3-5-sonnet or gemini-1.5-pro | Massive, highly accurate context retrieval |
| High-Volume Filtering & Quick Chat | gpt-4o-mini or llama-3.3-70b | Extremely low cost with sub-second response times |