Reflection AI open-sources Beam under Apache 2.0 with DFlash2 drafter for 3× faster decoding
Lianhuiq · x · 2026-10-17
Reflection AI has released Beam across platforms with open weights under the Apache 2.0 license, in collaboration with 40+ ecosystem partners. The release includes weights, tokenizer, and reference implementation; BF16, FP8, and NVFP4 weights plus a DFlash2 drafter; and Mirror CLI, the open-source harness it was trained on. Weights and model card are on Hugging Face. Collaborator zhijianliu notes DFlash 2 powers Beam from day one, decoding up to 3× faster with zero quality loss.
More from coding & agent
- TwelveLabs' video intelligence agent Jockey hits GA, plugs into ChatGPT and Claude via MCP — lmoroney · 2026-10-17
- Google demo: Gemini rebuilds ancient ruins into explorable 3D worlds via Antigravity + Blender MCP — jnack · 2026-10-17
- Mixpanel MCP Server Lets AI Agents Query Funnels and Retention via Natural Language — modelcontextprotocol · 2026-10-17
- MCP Connector Ties AI Assistants to UK's Largest FOI Platform WhatDoTheyKnow — modelcontextprotocol · 2026-10-17
- Codex mobile remote-control pain: server-relay design ignores at-home phone use — tinyfool · 2026-10-17
- Dev uses a custom Skill to cut a product launch video, extending AI agent workflows — AlchainHust · 2026-10-17