Beam 1.0 501B A23B
Neural Network
Beam 1.0 is an open-weight reasoning model from Reflection for code and long agent scenarios; Beam 1.0 API is available in BotHub.
Specifications
- Input
- text, documents
- Output
- text
- Context size
- 1 048 576 tokens
- Max answer length
- 943 718 tokens
- Reasoning effort
- low, medium, high, extra high, max
- Capabilities
- reasoning, function calling
Prompt cost
(per 1M tokens)
Answer cost
(per 1M tokens)
How it works Beam 1.0 501B A23B?
Beam 1.0 is an open-weight reasoning model from Reflection, built around two tasks: writing code and long agent scenarios where the goal must be maintained over many steps. The architecture is a sparse mixture of experts: 501 billion parameters in total and 23 billion active per token, meaning only a part of the network works for each request. In BotHub, the model is accessible from Russia without a VPN or foreign card; you pay in rubles for tokens as you go, not for a subscription, and you can switch between Beam 1.0 and other models in one window. For integrations, there is a unified OpenAI-compatible API: changing models does not require rewriting code, traffic is encrypted with AES-GCM, and companies have access to contracts, invoices, electronic document management, and an admin panel with limits. Developers will find Beam 1.0 useful for analyzing unfamiliar codebases, refactoring, and debugging; engineers for building agents that sequentially execute chains of tasks; analysts for step-by-step STEM problem breakdowns; and product teams for prototypes where multiple models work under one API.Frequently asked questions about Beam 1.0 501B A23B
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.
The model works with text and documents: a context of up to 1,048,576 tokens allows you to upload a large report, contract, or entire codebase. It is suitable for analyzing long files, writing and refactoring code, parsing logic, reasoning tasks, and for agents with tool calling.
The output limit is up to 943,718 tokens per response, which is hundreds of pages of text. It is enough for both a long document and a large code module in its entirety, without manual stitching of parts. In practice, the volume depends on your request.
Yes, the reasoning mode is confirmed: the model breaks down the task step-by-step before providing a final answer. This helps with math, logic, code debugging, and multi-step document analysis. Reasoning consumes tokens, so shorter requests are more cost-effective for simple questions.
Yes, function calling and structured JSON output are supported. You describe the schema and receive the answer in the required format or provide the model with a set of tools for an agent. In BotHub, this is available via a unified OpenAI-compatible API, so the model can be switched without rewriting the integration.
According to our data, text and documents are accepted as input — you can attach a file and ask to analyze its content. Images, audio, and video are not listed in the capabilities; for such tasks, there are specialized multimodal models in the same BotHub window.
Our data does not specify the model's working language, so we will not promise anything specific — it is easier to check with your own real request. If the result is not satisfactory, switch to another model in the same window and compare the answers.