Reflection Ai · 9× · tolest 07. Okt. 2026
67
Momentum
Beam is the first open-source language model from Reflection AI, unveiled on October 5, 2026. It is a sparse Mixture-of-Experts model with 501 billion total parameters and 23 billion active parameters per Token, designed for Coding, Reasoning, and agentic tasks. The model was pretrained on 23.8 trillion Tokens and features an effective context window of up to 1 million Tokens following Midtraining, though the current Beta-API is limited to 262,144 Tokens. The weights are scheduled to be released under the Apache-2.0 license in October 2026; currently, Beam is only accessible via a waitlist as a Beta-API, and no public pricing list exists yet.
Momentum-Verloop
09.07.07.10.
Features
| Key Benchmark (%) | SWE-Bench Verified: 80.9% |
| Context Window (Tokens) | 1,000,000 tokens effective (after midtraining); beta API currently 262,144 tokens |
| License | Apache 2.0 (planned for weights release in October 2026) |
| Multimodality | Text-only; no native image, audio, or file input |
| Platform | Beta API via platform.reflection.ai (waitlist), OpenAI-compatible; self-hosting only after weights release |
| Price per 1M Tokens | Not yet published (as of October 2026) |
| Release Date | Announced October 5, 2026; weights release later in October 2026 |
