As of: 2026-11-19 22:11 CET. Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/ Canonical page: https://postcutoff.com/m/reflection-beam/ # Beam (Beam-501B-A23B) Reflection AI, Beam family. Status: current. Type: reasoning model. Released: 2026-10-05. Training cutoff: 2026-06. ## How to call it | Provider | Model id | Endpoint | Docs | |---|---|---|---| | Reflection API (beta, early access / waitlist) | `Beam-501B-A23B` | https://api.reflection.ai/v1 | https://developers.reflection.ai/models | | Reflection API, OpenAI-compatible (Chat Completions, Models) | `Beam-501B-A23B` | https://api.reflection.ai/openai/v1 | https://developers.reflection.ai/ | | Hugging Face (open weights, Apache 2.0; BF16, FP8, NVFP4 and DFlash2 variants) | `reflection/Beam-1.0-501B-A23B` | | https://huggingface.co/reflection/Beam-1.0-501B-A23B | | OpenRouter (served by Baseten; price not listed at launch per VentureBeat) | | | | | Web app / early-access signup | | | https://platform.reflection.ai/ | ## Pricing - Price not published. - Last checked: 2026-10-17 ## Specs - Input: text - Output: text - Context window: 256,000 tokens - Max output: 128,000 tokens - Open weights: yes - Licence: apache-2.0 ## What it has not seen - Published training cutoff: 2026-06 (June 2026). - AI events after it on Post-Cutoff: 1,923 (451 major, 84 historic), as of 2026-11-19. - Already missing at launch: 686. - Briefings of the news since June 2026: https://postcutoff.com/briefings/cutoff-2026-06-s.md (short), https://postcutoff.com/briefings/cutoff-2026-06-m.md (medium), https://postcutoff.com/briefings/cutoff-2026-06.md (full). - Its card, with five things it has never heard of: https://postcutoff.com/you/reflection-beam/ ## What stands out - Efficient 500B-class open MoE: 501B total / 23B active sparse MoE trained on 23.8T tokens; Reflection says it matches GLM-5.2 on reasoning with 3-4x less inference compute. Source: https://reflection.ai/blog/introducing-beam - Large-scale agentic RL: RL on 10.5K GB300 GPUs for 4 weeks: >100M rollouts across ~1M coding, agentic and STEM environments using ~1.3B sandboxes. Source: https://reflection.ai/blog/introducing-beam - 1M-token context (training): Pretrained at 256K and extended to 1M tokens in midtraining; the beta API currently exposes 256K context and 128K output. Source: https://developers.reflection.ai/models ## Notes Announced Oct 5, 2026; Apache-2.0 weights released on Hugging Face Oct 16, 2026 (reflection/Beam-1.0-501B-A23B, plus FP8/NVFP4 quantizations; 8-bit and 4-bit versions fit one 8-GPU server per Reflection), with a final model card and the Beam Safety Technical Report. Reflection hosted API free during public beta. API is beta, reasoning always on (default effort medium), tool calling and structured outputs. No published price. Benchmarks are Reflection's own: SWE-bench Verified 80.9, Terminal Bench v2.1 80.1, GPQA Diamond 90.5, HLE no tools 36.2. Reflection AI's first model. Until the weights ship, access is through the waitlisted beta API. The OpenAI-compatible endpoint works with the official OpenAI SDKs once you change `base_url` to `https://api.reflection.ai/openai/v1` and use a Reflection API key. When the weights are released, add the Hugging Face repo and switch `status` to `current`. Model page: https://postcutoff.com/m/reflection-beam/