Inference
EN
Reflection AI

Beam 1.0 501B-A23B

Beam 1.0 501B-A23B is a reasoning model from Reflection AI in the Beam family. 27 benchmarks count toward its index, in 6 categories.

  • Open weights
  • Reasoning
  • Input: text
  • Released 2026-10
RankedEnough results for an index.
IndexAll categories, out of 100.60.2±4.6Median 51.0
CoverageIndex weight that has results.85%
SpeedOutput tokens per second.345/sMedian 67/s
First tokenSeconds to the first output token.6.4 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.—Median $0.60
Output / 1MUS dollars per 1M output tokens.—Median $2.55
ContextMaximum tokens in one request.1MMedian 262K
EloLMArena rating and rank.—Median 1422

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.15 counted59.376 / 218
CodingCode writing and repair.3 counted62.058 / 209
ReasoningLogic problems and puzzles.4 counted63.253 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.3 counted56.583 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted63.4—
MathMath problems.1 counted55.6—

Results

27 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
AIME 2026Math97.7%3 / 34
Software Engineering Benchmark VerifiedCoding81.6%12 / 114
LongBench v2Reasoning67.5%3 / 25
SWE-bench ProCoding66.1%12 / 83
Critical Physics TasksReasoning16.3%47 / 214
Scientific Code BenchmarkCoding49.2%8 / 35
Instruction Following BenchmarkInstruction79.6%14 / 57
MCP AtlasAgentic80.8%14 / 52
Artificial Analysis Humanity's Last ExamKnowledge34.7%64 / 222
GPQA Diamond · BenchLMKnowledge90.6%25 / 84
Artificial Analysis Long Context ReasoningReasoning78.2%68 / 217
GDPval-AA normalizedAgentic33.9%54 / 133
Terminal-Bench 2.1 · BenchLMAgentic80.1%36 / 86
Artificial Analysis Omniscience AccuracyKnowledge30.7%77 / 164
Artificial Analysis CyberGym-E2E-AAAgenticAbove fitFar above this model's index.45.0%9 / 19
Artificial Analysis SciCodeCoding48.5%57 / 114
BrowseCompAgentic76.8%35 / 63
Humanity's Last Exam without toolsKnowledge36.0%30 / 53
DeepSearchQAAgentic80.5%14 / 24
Artificial Analysis Cyber IndexAgentic33.7%12 / 19
τ³-Bench Tool-Agent-User EvaluationAgentic36.1%21 / 32
Humanity's Last Exam with toolsAgentic51.1%25 / 36
Medical Long Context Reasoning (MLCR-AA)Reasoning17.8%17 / 22
Artificial Analysis AutomationBenchAgentic48.5%18 / 23
Artificial Analysis CWE-Bench-AAAgentic49.2%16 / 19
Artificial Analysis DeepsecBench-AAAgentic6.9%19 / 21
DeepSWEAgentic43.3%54 / 59
Artificial Analysis GDP.pdfAgentic11.6%21 / 22
Artificial Analysis Terminal-Bench v4.0Agentic3.0%22 / 22
SWE-Atlas Codebase QnACoding31.8%—

Sources

BenchLMCC BY-NC 4.0

History

5 releases
Index59.9 → 60.2
23 Oct 202612 Nov 2026

Questions

How good is Beam 1.0 501B-A23B?
Its index is 60.2 out of 100. That is place 72 of 283 ranked models.
What is Beam 1.0 501B-A23B best at?
Reasoning: 63.2 out of 100, place 53 of 227.
How fast is Beam 1.0 501B-A23B?
345 output tokens per second, 6.4 s to the first token.
What is the context window of Beam 1.0 501B-A23B?
1M tokens.
Is Beam 1.0 501B-A23B open source?
Yes. Its weights are published.
When was Beam 1.0 501B-A23B released?
October 2026.

Badge

Beam 1.0 501B-A23B on Inference360: place 72 of 283, index 60.2
HTML
<a href="https://inference360.ai/models/reflection-beam"><img src="https://inference360.ai/badges/reflection-beam.svg" alt="Beam 1.0 501B-A23B on Inference360: place 72 of 283, index 60.2" height="20"></a>
Markdown
[![Beam 1.0 501B-A23B on Inference360: place 72 of 283, index 60.2](https://inference360.ai/badges/reflection-beam.svg)](https://inference360.ai/models/reflection-beam)

Compare with

Claude Opus 4.6 (Adaptive)60.0Sage-A1.560.5Beam 1.0 501B-A23B FP859.2

More from Reflection AI

All models
Beam 1.0 501B-A23B FP859.2Beam 1.0 501B-A23B NVFP458.5