Anthropic's most capable model for the hardest reasoning and long-horizon agentic work; thinking always on, 1M context, up to 128K output tokens.
Input
2,711,700
Output
13,558,500
Cache (read)
67,793
Cache (write)
3,389,625
claude-fable-5-1
AI models and pricing
Access more than a hundred language, vision and audio models from OpenAI, Anthropic, Google and others through a single API key, paid in Toman and with no foreign bank account. Point the base URL of any OpenAI-compatible client at CupAI and you reach every model without changing a line of code, with all spending managed from one wallet.
Prices track each provider's own rate and are converted using the live exchange rate, and every request is billed from your wallet. Text models are billed on their input and output tokens; video models on the seconds they produce, and image models per generated image. Retired models are clearly marked and carry a suggested replacement, so moving to a newer version is straightforward.
233 models
GPT-6 Astra
Over 272KOpenAI's new flagship reasoning model for long-horizon agentic work: computer use, coding and multi-step research; 1.05M context, up to 128K output tokens.
Input
2,711,700
Output
13,558,500
Cache (read)
271,170
Cache (write)
3,389,625
gpt-6-astra
GPT-6.1 Sol
Over 272KOpenAI's reasoning model that nearly matches GPT-6 Astra at a fifth of the price; 1.05M context, up to 128K output tokens.
Input
542,340
Output
2,711,700
Cache (read)
27,117
Cache (write)
677,925
gpt-6.1-sol
Strongest Claude Opus model for coding, agents, and professional work
Input
1,355,850
Output
6,779,250
Cache (read)
1,355,850
Cache (write)
1,694,813
claude-opus-5
Anthropic's Opus model for long-running agentic coding and knowledge work
Input
1,084,680
Output
5,423,400
Cache (read)
54,234
Cache (write)
1,355,850
claude-opus-5-5
Reasoning-first Gemini preview for agentic coding and complex problem solving
Input
1,084,680
Output
4,881,060
Cache (read)
1,084,680
Cache (write)
1,084,680
same as input
gemini-3.1-pro-preview
Providers
OpenAI
The research lab behind the GPT, o-series, DALL·E, Whisper and Sora families, and the originator of the Chat Completions and Responses APIs most of this catalogue speaks.
Anthropic
Maker of the Claude family — Opus, Sonnet and Haiku — known for long context windows, extended thinking and strong coding and agentic tool use.
Google DeepMind
Google's AI division and the source of Gemini, Imagen and Veo: natively multimodal models with very large context windows, served over the Gemini API.
How model pricing works
Every AI model strikes a different balance between speed, answer quality and cost. Flagship models suit complex reasoning, long-document analysis and precise code generation, while lighter models cost considerably less for summarising, classification and short answers. The list above shows each model's rate broken down by input token, output token and cache, so you can estimate the cost of your workload before you start.
A token is the smallest unit of text a model processes; roughly speaking a thousand tokens is about 750 English words. The cost of a request is the sum of its input and output tokens, so shortening your prompt and capping the response length reduces cost directly. Some models have a second price tier for very long inputs, which reprices the entire request once it crosses a stated threshold.
Frequently asked questions
- How is usage billed?
- By the input and output tokens of each request, deducted from your wallet in Toman. Every model's rate is on this page.
- Are the prices the same as the provider's?
- Yes. The base price is the provider's own rate, converted to Toman at the live exchange rate.
- Which model should I start with?
- A fast mid-range model is enough for general work. If you need high accuracy or your input is very long, use one of the flagship models.
- Do OpenAI-compatible clients work?
- Yes. Just point the base URL at CupAI and use your own API key.