Beam 1.0 501B A23B

Reflection's Apache 2.0 open-weight text model with 501.3B total and 23.4B active parameters, built for coding, reasoning and tool use. It supports a shared 1,048,576-token context window, up to 131,072 output tokens, JSON mode and structured output. Reasoning is always enabled with low, medium, high, xhigh and max effort; NanoGPT defaults to max. Knowledge cutoff: June 30, 2026.

Reasoning
Tool Calling
Structured Output

Added Oct 16, 2026

Pricing

Auto routing · per 1M tokens

Input: $0.50
Output: $1.50
Cache read: $0.050

Compare provider prices

Specifications

Context window: 1M
Max output: 131.1K
Parameters: 501.3B / 23.4B; Total / active
Avg output (7d): 1.8K tokens; Longer than 88% of models

Benchmarks

Artificial Analysis

Design Arena

Sourced from Artificial Analysis.

Intelligence Index

30.3

Better than 84% of models compared

Agentic work

AutomationBench-AA

Workflow automation with guardrail penalties

48.5%

Better than 64% of models compared

AutomationBench-AA Tasks Completed

Fully completed workflows without guardrail violations

17.0%

Better than 63% of models compared

AA-Briefcase v1.1

Agentic knowledge work (Elo)

976 Elo

Better than 45% of models compared

GDPval-AA v2.1

Economically valuable tasks (Elo)

1177 Elo

Better than 58% of models compared

Document reasoning

GDP.pdf

Professional PDF reasoning: all-pass rate

11.6%

Better than 48% of models compared

AA-LCR v1.1

Long context reasoning with updated grading

78.0%

Better than 78% of models compared

Reasoning

HLE

Humanity's Last Exam

34.7%

Better than 81% of models compared

Coding

Terminal-Bench v4.0

Practical coding and terminal tasks

3.0%

Better than 51% of models compared

SciCode

Python programming for scientific computing

48.5%

Better than 47% of models compared

Legacy benchmarks

AA-LCR (unversioned / legacy)

Long context reasoning evaluation