Reflection
Beam 1.0 501B A23B
Sparse mixture-of-experts reasoning model with 23B active parameters for coding and agentic workflows
reflection/beam-1.0-501b-a23b
Context
1M
Output
131.1K
Released
2026-10-05
Knowledge cutoff
2026-06-30
Types
InputText
OutputText
Reasoning
Supported
Tool use
Supported
Structured output
Supported
Attachments
—
Open weights
Supported
Provider pricing
Kilo Gateway
InText
OutText
- Price
- $0.25 / $1.00
- Context
- 1M
- Status
- Stable
OpenRouter
InText
OutText
- Price
- $1.00 / $3.00
- Context
- 1M
- Status
- Stable
Benchmarks
DeepSWE
43.3
SWE-Bench Pro
76.5
SWE-Bench Pro
66.1
Terminal-Bench
80.1
SWE-Atlas Codebase QnA
31.8
SWE-Bench Multilingual
78.3
SWE-Bench Verified
81.6
AIME 2026
97.7
Humanity's Last Exam
36
Humanity's Last Exam
51.1
SciCode
49.2
CritPt
16.3
GPQA Diamond
90.6
LiveCodeBench
88.3
AutomationBench
39.3
MCP Atlas
80.8
τ3 Banking
36.1
BrowseComp
76.8
DeepSearchQA
80.5
AA-LCR
78.2
LongBench
67.5
IFBench
79.6
AA-Omniscience
10.8 · index