Reflection AI's Beam model now available on AI Gateway for free

Reflection AI has released Beam, a low-latency US open-weight reasoning model designed for coding and agentic tasks, now available on AI Gateway. Previously known in stealth as stealth/cat-jam, Beam is free to use for a limited time during the launch period. The model's weights are released under the Apache 2.0 license.

Model Architecture and Capabilities

Beam supports tool calling and structured outputs with a 1M-token context window for code, documents, and conversation history. It is optimized for speed by activating only 23B of its 501B parameters per token and avoiding unnecessary tokens, allowing it to reach answers faster than larger open models. Reasoning effort is adjustable to suit specific tasks.

Access and Integration

Users can access Beam for free via two methods on AI Gateway:

• Using the standard model name reflection/beam, which automatically transitions to paid pricing when the free period ends.
• Using the -free suffix reflection/beam-free, which stops serving the model when the free offer expires.

To integrate Beam with coding agents like Claude Code, Codex, and Cursor, users must install the latest Vercel CLI and run setup, then select the appropriate model slug. AI Gateway provides a unified API for model calls, usage tracking, and performance optimizations, reflecting provider pricing with no markup or platform fees.