Beam: Another Open Model Joins The Frontier Race

Beam joins the open model race: Reflection's Apache 2.0 model scores 18 on Artificial Analysis, writes roughly 3× fewer tokens than DeepSeek, but ranks second among US-built open models, not the frontier. Reflection released Beam under Apache 2.0 on October 16, 2026, and independent benchmarks immediately showed the trade-off: on Artificial Analysis's Intelligence Index, Beam scores 18, well below the 23 median of comparable open models and behind DeepSeek V4.1 Flash's 32. The fewer-tokens claim held up, Beam generated 54 million tokens versus 170 million for DeepSeek on the same test, but solution quality did not match. Each token costs less to produce because Beam routes through only 8 of its 256 expert blocks, and inference runs about 300 tokens a second; evaluating it costs 28 cents per Artificial Analysis task versus 60 for DeepSeek. As a coding agent in Mirror, Beam scored 28 on the Coding Agent Index, 40 points behind the leading setups. Running it yourself requires four workstation GPUs or a Mac with streaming code not yet merged; the four-bit file is 307 gigabytes. On Vals and Epoch, Beam ranks 29th of 32 and 60th of 254 respectively. If you need an open model built in the US with no commercial restrictions, Beam is the top choice on Artificial Analysis's list, but for serious work, DeepSeek V4.1 Flash remains the standard. Chapters: 0:00 Intro 0:58 Where It Ranks 2:02 Fewer Tokens 4:14 On Real Work 5:41 Running It 7:44 Verdict Tools & resources mentioned: - Beam: https://huggingface.co/reflection/beam - Mirror: https://github.com/MirrorNetworking/Mirror - llama.cpp: https://github.com/ggml-org/llama.cpp - vLLM: https://github.com/vllm-project/vllm About The Stack The Stack helps you build with AI. Each video takes one tool, model, or workflow and shows how it works in a few focused minutes, with the real benchmarks and real costs. We go deep on Claude Code and Cursor for AI coding, AI agents and MCP servers, the open-source AI tools and GitHub repos most people miss, RAG and vector search, fine-tuning, and running local LLMs on your own machine with Ollama and LM Studio. We compare models like ChatGPT and Claude, test AI automation with Zapier, Make, and n8n, and flag the tools that actually ship. Subscribe for new breakdowns: https://www.youtube.com/@the-stack-ai?sub_confirmation=1 #beam ai model #ai news #open models