Reflection AI's Beam open-weight model lands on Vercel AI Gateway for free
Vercel has added Beam, a reasoning model from Reflection AI, to its AI Gateway, and developers can use it for free for a limited time during the launch period. Beam is a low latency US open-weight model built for coding and agentic tasks, with weights released under Apache 2.0, according to Vercel.
- Reflection AI's Beam, previously in stealth as stealth/cat-jam, is now on Vercel's AI Gateway and free for a limited time during launch.
- Beam is a low latency US open-weight reasoning model for coding and agentic tasks, with weights under Apache 2.0.
- It has a 1M-token context window, tool calling, structured outputs and adjustable reasoning effort.
- According to Vercel, it activates 23B of 501B parameters per token.
- reflection/beam moves to paid pricing automatically after the free period, while reflection/beam-free stops serving.
What Beam offers
Beam was previously available in stealth under the name stealth/cat-jam. Vercel describes it as a low latency US open-weight reasoning model aimed at coding and agentic tasks. Its weights are released under the Apache 2.0 license.
The model supports tool calling and structured outputs. It has a 1M-token context window, which Vercel says is meant for code, documents and conversation history. Reasoning effort is adjustable, so developers can tune how much thinking the model does for a given task.
Speed is the main design goal. According to Vercel, Beam activates only 23B of its 501B parameters per token and avoids producing tokens a task does not need. As a result, it tends to reach answers sooner than larger open models, the company says.
Two ways to use it for free
Vercel lists two routes to the free offer, and they behave differently once the promotion ends.
- Standard model name: reflection/beam. When the free period ends, this slug will automatically move to paid pricing with no downtime.
- The -free suffix: reflection/beam-free. This model will stop serving when the free offer ends.
The first option suits teams that want to keep using Beam without changing their code after the launch period. The second works as a hard stop, since the model simply stops serving at the end of the offer.
In the AI SDK, a call needs only the model string set to reflection/beam inside a streamText request. Vercel also provides a Chat Completions option and a model playground for trying Beam.
Setup for coding agents
Vercel says Beam can be used in Claude Code, Codex, Cursor and other agents. Developers need to install the latest Vercel CLI and run npx vercel ai-gateway setup. They then select reflection/beam or reflection/beam-free inside the agent. Vercel points to its coding agents guide for details.
About AI Gateway
AI Gateway offers a unified API for calling models, tracking usage and cost, and configuring retries, failover and performance optimizations. Vercel says this is aimed at higher-than-provider uptime. It also includes custom reporting, Zero Data Retention support, budgets for API keys and routing rules.
Vercel states that AI Gateway reflects provider pricing with no markup and charges no platform fee on inference, including on Bring Your Own Key (BYOK) requests.
What it means for developers
For developers, the free launch period is a low-cost chance to test an open-weight reasoning model on coding and agent workloads through a single API. The long context window and adjustable reasoning effort give room to experiment with larger codebases and different speed and depth trade-offs. Vercel has not stated in the source how long the free period will last or what Beam will cost afterward, so teams planning longer-term use should check pricing in the model details once the offer ends.
