Mercury 2
Neural Network
Mercury 2 is a diffusion reasoning language model: parallel token generation instead of sequential, available via API.
Specifications
- Input
- text, images, documents
- Output
- text
- Context size
- 128 000 tokens
- Max answer length
- 50 000 tokens
- Reasoning effort
- none, low, medium, high
- Capabilities
- reasoning, function calling
Prompt cost
(per 1M tokens)
Answer cost
(per 1M tokens)
How it works Mercury 2?
Mercury 2 is a reasoning language model built on a diffusion approach: instead of outputting one token at a time, it forms and refines multiple tokens in parallel, resulting in faster responses. It is the first reasoning model of this class, designed for tasks where both response logic and latency are critical. Via BotHub, Mercury 2 is available without VPN or foreign cards, with pay-as-you-go billing in rubles, no subscription required, and unused tokens do not expire. Access over 250 models in one window to easily compare answers without changing the interface. A unified OpenAI-compatible API allows you to connect the model to your service and switch to another without rewriting the integration. Companies get access to contracts, invoices, EDI, and an admin panel with limits; chats are protected by AES-GCM encryption. It is useful if you are building an interactive assistant where the user expects real-time responses, solving analytical tasks with step-by-step reasoning, processing large volumes of similar requests, or building agent chains where the speed of each iteration multiplies.Frequently asked questions about Mercury 2
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
This is a text model by Inception with a 128,000 token context. It accepts not only text but also images and documents, can reason before answering, and call external functions. Suitable for coding, parsing long files, analytics, and STEM tasks.
The single response limit is up to 50,000 tokens, with a total context window of 128,000 tokens. This is sufficient for a large code module, a detailed technical analysis, or an extensive document that doesn't need to be assembled from multiple separate fragments.
Yes, the reasoning mode is confirmed by our data: the model can build intermediate steps before producing a result. This significantly helps where output sequence is important — mathematics, logic, code debugging, requirement analysis, and comparing multiple solution options.
Yes, the model supports function calling and structured JSON output, making it easy to integrate into agents and pipelines. On BotHub, it is available via a unified OpenAI-compatible API: you can switch scenarios to another model without rewriting the integration.
The model accepts images and documents as input — this is confirmed by our data, so you can analyze screenshots, diagrams, and files. We have no notes regarding audio and video, so please test this scenario in advance if you plan to build a workflow around it.
We do not have separate confirmed measurements for the Russian language, so it is more reliable to test the model on your own task. In BotHub, this is done in one window: send a request to Mercury 2 and compare the answer with other models. Access from Russia, payment by ruble card, no VPN needed.