Beam 1.0 added to BenchGecko · Reflection 501B open-weight MoE with 23B active and 1M context
Reflection released Beam 1.0 open weights on 2026-10-16 under Apache 2.0. BenchGecko lists it at $0.25 input and $1.00 output per M tokens with a 1,048,576 token context window.
What launched
Reflection released Beam 1.0 on 2026-10-16 under Apache 2.0. The open-weight checkpoint is Beam-1.0-501B-A23B on Hugging Face, with BF16, FP8 and NVFP4 formats. It is Reflection's first public model and targets coding and long-horizon agentic workflows.
BenchGecko added the model today with OpenRouter pricing from the Nebius fp8 endpoint.
Specs
From Reflection and Hugging Face listings:
- Architecture · sparse Mixture-of-Experts, 501.3B total parameters, 23.4B active per token
- Experts · 256 per MoE layer, 8 routed per token
- Depth · 52 layers
- Context · 1M tokens, stored as 1,048,576 on BenchGecko to match provider listings
- Inputs · text
- Output · text
- Reasoning effort · low, medium (default), high, xhigh or max
- Day-0 runtimes · transformers, vLLM and SGLang
Licence
Apache 2.0. Commercial use, modification and redistribution are allowed under that licence.
Pricing on BenchGecko
OpenRouter lists reflection/beam-1.0-501b-a23b at $0.25 per M input tokens and $1.00 per M output tokens. As of publish time, the live OpenRouter endpoints response shows one paid route: Nebius at fp8 quantization with 1,048,576 context and 131,072 max completion tokens.
BenchGecko stores those figures on the model row as pricing_input 0.25 and pricing_output 1.0.
Vendor-reported benchmarks
Reflection reports these BF16 results. BenchGecko records them as vendor claims and has no Gecko Tests for this model yet.
- SWE-bench Verified · 81.6
- Terminal-Bench v2.1 · 80.1
- GPQA Diamond · 90.6
- AIME 2026 · 97.7
- HLE, no tools · 36.0
- HLE, with tools · 51.1
- MCP Atlas · 80.8
Sources
Primary sources used for this listing:
Source: BenchGecko
Written automatically from BenchGecko's tracked data for this period; every number comes from the changelog, model, mindshare and pricing datasets. Checked against the source data before publishing.