# Reflection 开源权重模型 Beam 上架 Ollama,限时免费可用
> 原标题:Reflection's Beam is freely available on Ollama for a limited time
- 来源:Ollama:Blog
- 发布时间:2026-10-16T00:00:00.000Z
- AIHOT:https://aihot.news/items/f14brdupsvks0k2hoang4lx6w
- 原文:https://ollama.com/blog/reflection-beam
## 摘要
Reflection 首个开源权重模型 Beam 发布,Ollama 作为 day-zero 合作方在 Ollama Cloud 上限时免费提供,且全程零数据保留(ZDR)。
## 推荐理由
Ollama 作为首发合作方介绍 Beam 的参数结构、上下文窗口和 ZDR 云端接入方式,读者可以据此评估在本机或云端运行该模型的可行性。
## 正文 · 中文译文
## 2026年10月16日
Reflection 的首个开放权重模型 Beam 于今天上午正式发布,我们是零日合作伙伴,在 Ollama 上免费推出(仅限时!)。与 Reflection 合作,Ollama 的 Beam 端点运行时完全零数据保留(ZDR)。你可以通过 [Ollama Cloud](https://docs.ollama.com/cloud) 访问 Beam,在那里你可以从应用或终端运行支持的模型,无需下载模型或管理本地推理。
Beam 是一个混合专家模型,总参数量约为 5010 亿,每个 token 激活约 230 亿参数,专为软件工程、推理和长时程智能体任务而设计。
Beam 专为不止于单次响应的工作流而构建:编写和维护代码、跨多个步骤使用工具,以及随时间推移处理复杂任务。其文本输入、文本输出的设计聚焦于这些工作负载,其中代码、工具调用、日志、文档和结构化输出是核心。
Beam 拥有 100 万 token 上下文窗口,以支持长时间运行的工作流。
Beam 以 Apache 2.0 许可发布,让开发者可以自由检查、定制并基于该模型进行构建。
## 你可以用 Beam 构建什么
Beam 专为以下工作负载而设计:
- **编码智能体:**解决软件工程任务、处理代码库,并帮助编写和维护代码。
- **长时间运行的智能体:**执行多步骤工作流、使用工具,并在延长的操作序列中朝着目标推进。
- **数学与科学推理:**处理复杂的数学和科学问题。
- **自定义模型:**针对特定应用微调和定制开放权重模型。
## 基准测试




## 开始使用
下载 [最新版本的 Ollama](https://ollama.com/download) 以开始使用 Beam。
```
ollama run beam:cloud
```
Claude Code
```
ollama launch claude --model beam:cloud
```
Codex
```
ollama launch codex --model beam:cloud
```
OpenCode
```
ollama launch opencode --model beam:cloud
```
查看模型页面了解更多信息:
- [Beam 模型页面](https://ollama.com/library/beam)
## 更多信息
- [Reflection 的 Beam 公告](https://reflection.ai/blog/beam-is-available)
- [Reflection Beam 模型卡](https://reflection.ai/beam-model-card)
- [Ollama 集成](https://docs.ollama.com/integrations)
- [下载 Ollama](https://ollama.com/download)
## 正文 · 原文
## October 16, 2026
Reflection’s first open-weight model Beam is generally available as of this morning, and we’re day-zero partners rolling it out on Ollama for free (for a limited time only!). In collaboration with Reflection, Ollama’s Beam endpoint runs with full zero data retention (ZDR). You can access Beam through [Ollama Cloud](https://docs.ollama.com/cloud), where you can run supported models from your apps or terminal without downloading models or managing local inference.
Beam is a mixture-of-experts model with approximately 501 billion total parameters and 23 billion active parameters per token, designed for software engineering, reasoning, and long-horizon agentic tasks.
Beam is built for workflows that involve more than a single response: writing and maintaining code, using tools across multiple steps, and working through complex tasks over time. Its text-in, text-out design focuses on these workloads, where code, tool calls, logs, documents, and structured outputs are central.
Beam has 1 million token context window to support long-running workflows.
Beam is released under Apache 2.0, giving developers the freedom to inspect, customize, and build on the model.
## What you can build with Beam
Beam is designed for workloads such as:
- **Coding agents:** Solve software engineering tasks, work through codebases, and help write and maintain code.
- **Long-running agents:** Execute multi-step workflows, use tools, and work toward a goal over an extended sequence of actions.
- **Math and science reasoning:** Work through complex mathematical and scientific problems.
- **Custom models:** Fine-tune and customize an open-weight model for specific applications.
## Benchmarks




## Get started
Download the [latest version of Ollama](https://ollama.com/download) to get started with Beam.
```
ollama run beam:cloud
```
Claude Code
```
ollama launch claude --model beam:cloud
```
Codex
```
ollama launch codex --model beam:cloud
```
OpenCode
```
ollama launch opencode --model beam:cloud
```
See the model page for more information:
- [Beam model page](https://ollama.com/library/beam)
## More information
- [Reflection’s Beam announcement](https://reflection.ai/blog/beam-is-available)
- [Reflection Beam model card](https://reflection.ai/beam-model-card)
- [Ollama integrations](https://docs.ollama.com/integrations)
- [Download Ollama](https://ollama.com/download)