Beam 1.0

Reflection AI

Beam 1.0 is Reflection AI's first open-weight model: a sparse Mixture-of-Experts reasoning model with 501.3B total parameters and 23.4B active per token (52 layers: 3 dense + 49 MoE; 256 experts; gated GQA with sliding-window attention; 152K vocabulary), pretrained on 23.8T tokens and post-trained with large-scale asynchronous RL. It targets coding, long-horizon agentic work, and tool use, with a 1M-token context window, up to 128K output tokens on the Reflection API, text-only input/output, and mandatory reasoning with low/medium/high/xhigh/max effort levels. Released under Apache 2.0 in BF16, FP8, and NVFP4 with a DFlash speculative-decoding drafter; available on Hugging Face, the Reflection API (Beam-1.0-501B-A23B), OpenRouter, and other partners.

Key Specifications

Parameters
501.3B
Context
1.0M
Release Date
October 16, 2026
Average Score
61.6%

Timeline

Key dates in the model's history
Announcement
October 5, 2026
Release
October 16, 2026
Last Update
October 27, 2026
Today
November 20, 2026

Technical Specifications

Parameters
501.3B
Training Tokens
23.8T tokens
Knowledge Cutoff
June 30, 2026
Family
-
Capabilities
MultimodalZeroEval

Pricing & Availability

Input (per 1M tokens)
$0.40
Output (per 1M tokens)
$1.20
Max Input Tokens
1.0M
Max Output Tokens
131.1K
Supported Features
Function CallingStructured OutputCode ExecutionWeb SearchBatch InferenceFine-tuning

Benchmark Results

Model performance metrics across various tests and benchmarks

Programming

Programming skills tests
SWE-Bench Verified
Reflection model card; Beam BF16; temp=0.7, top_p=0.99 • Self-reported
81.6%

Reasoning

Logical reasoning and analysis
GPQA
Reflection model card; Beam BF16; temp=0.7, top_p=0.99; GPQA Diamond • Self-reported
90.6%

Other Tests

Specialized benchmarks
AA-LCR
Reflection model card; Beam BF16; temp=0.7, top_p=0.99 • Self-reported
78.2%
AA-Omniscience Index
Reflection model card; Beam BF16; temp=0.7, top_p=0.99; public split only, Reflection's own run • Self-reported
5.4%
AIME 2026
Reflection model card; Beam BF16; temp=0.7, top_p=0.99 • Self-reported
97.7%
AutomationBench
Reflection model card; Beam BF16; temp=0.7, top_p=0.99; public split only • Self-reported
39.3%
BrowseComp
Reflection model card; Beam BF16; temp=0.7, top_p=0.99; with context management • Self-reported
76.8%
CritPT
Reflection model card; Beam BF16; temp=0.7, top_p=0.99; Artificial Analysis harness • Self-reported
16.3%
DeepSearchQA
Reflection model card; Beam BF16; temp=0.7, top_p=0.99; with context management • Self-reported
80.5%
DeepSWE 1.1
Reflection model card; Beam BF16; temp=0.7, top_p=0.99 • Self-reported
43.3%
Humanity's Last Exam
Reflection model card; Beam BF16; temp=0.7, top_p=0.99; no tools • Self-reported
36.0%
Humanity's Last Exam (with tools)
Reflection model card; Beam BF16; temp=0.7, top_p=0.99; with tools • Self-reported
51.1%
IFBench
Reflection model card; Beam BF16; temp=0.7, top_p=0.99 • Self-reported
79.6%
LiveCodeBench
Reflection model card; Beam BF16; temp=0.7, top_p=0.99 • Self-reported
88.3%
LongBench v2
Reflection model card; Beam BF16; temp=0.7, top_p=0.99 • Self-reported
67.5%
MCP Atlas
Reflection model card; Beam BF16; temp=0.7, top_p=0.99 • Self-reported
80.8%
SciCode
Reflection model card; Beam BF16; temp=0.7, top_p=0.99 • Self-reported
49.2%
SWE Atlas - Codebase QnA
Reflection model card; Beam BF16; temp=0.7, top_p=0.99 • Self-reported
31.8%
SWE-bench Multilingual
Reflection model card; Beam BF16; temp=0.7, top_p=0.99 • Self-reported
78.3%
SWE-Bench Pro
Reflection model card; Beam BF16; temp=0.7, top_p=0.99; SWE-Bench Pro v1 public set • Self-reported
66.1%
Tau3 Banking
Reflection model card; Beam BF16; temp=0.7, top_p=0.99 • Self-reported
36.1%
Terminal-Bench 2.1
Reflection model card; Beam BF16; temp=0.7, top_p=0.99 • Self-reported
80.1%

License & Metadata

License
apache_2_0
Announcement Date
October 5, 2026
Last Updated
October 27, 2026

Compare Beam 1.0

All comparisons

Similar Models

All Models

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.