Skip to main content
BenchLM
Data

Data as of November 19, 2026 · How the score is built

Beam 1.0 501B-A23B

CurrentOpen WeightReasoning

Decision readingBeam 1.0 501B-A23B scores 41.2 out of 100 and ranks #132 of 214. This profile shows 22 source-displayable benchmark rows; its strongest eligible category is Agentic at #74. No comparable first-party API price is published in the catalog.

Email me when its price, rank or availability changes.

Capability

41.2/100

field median 49#132 of 214 ranked models

Public

#132of 214

Verified —

Price

API rate not published

input median $0.60No comparable first-party hosted token rate

Speed

273tok/s

field median 93 tok/sFirst token 7.73 s

Context

1Mtokens

field median 256,000Maximum output length is tracked separately

Decision snapshot

Where this model sits in the ranked field, its strongest published evidence, and what to validate before choosing it. The marker compares it with the current ranked catalog, not with an absolute quality threshold.

Strongest published evidence

Agentic ranks #74. Particularly useful for coding agents, browser research, and computer-use workflows.

Validate before choosing

22 published rows leave some tracked benchmark slots empty. No comparable first-party API token rate is published.

Source-linked · 22 displayable benchmark rows

Is Beam 1.0 501B-A23B right for your job?

Its place on the LLM Selector’s job rankings, the same order the job pages publish. Describe your own job to see where it stands for that.

Describe the task and constraints. This selector recommends models; it cannot carry out the task or access your files and accounts.

We keep what you type here, drafts included, for 180 days, and after that only the job and requirements read from it. Nothing is tied to your IP address or account. Leave out names and other personal details. Privacy

Category score record

Scores and ranks appear only where published evidence can be displayed. The table keeps the score, weight, cohort, and evidence state together.

Category scores, ranks, weighting, benchmark coverage, and evidence status
CategoryScore
AgenticRank #74 of 138Percentile 47thWeight 22%7 benchmarksVerified
40.8
CodingRank #93 of 160Percentile 42ndWeight 20%8 benchmarksVerified
32.5
ReasoningRank #99 of 159Percentile 38thWeight 17%2 benchmarksVerified
37.4
MultimodalWeight 12%0 benchmarksNot measured
Not measured
KnowledgeRank #92 of 179Percentile 49thWeight 12%2 benchmarksVerified
46.2
MultilingualWeight 7%0 benchmarksNot measured
Not measured
Inst. FollowingRank Not rankedWeight 5%1 benchmarkVerified
88.0
MathRank Not rankedWeight 5%1 benchmarkVerified
84.6

22 of 923 tracked benchmark slots have displayable evidence · bars run 0–100

Coverage details

How much of this is verified

Coverage is split by category so a strong number never hides a thin evidence base. Verified means the row is tied to a published source; provisional rows remain visible but separate.

  1. Agentic7/7 verified
  2. Coding8/8 verified
  3. Reasoning2/2 verified
  4. MultimodalNot measured
  5. Knowledge2/2 verified
  6. MultilingualNot measured
  7. Inst. Following1/1 verified
  8. Math1/1 verified
Verified sourceProvisionalNot measured

Capability shape

Each axis shows percentile within that category’s eligible cohort. The comparison outline is the median of the six nearest public-score peers; a collapsed vertex means the category is not rank-eligible.

Beam 1.0 501B-A23B category percentile values

  • Agentic47th percentile
  • Coding42nd percentile
  • Reasoning38th percentile
  • MultimodalNot eligible
  • Knowledge49th percentile
  • MultilingualNot eligible
  • Instruction followingNot eligible
  • MathNot eligible

The dashed outline is median of 6 nearest peers.

Eligible category ranks

  1. Agentic#74/138
  2. Coding#93/160
  3. Reasoning#99/159
  4. MultimodalNot ranked
  5. Knowledge#92/179
  6. MultilingualNot ranked
  7. Inst. FollowingNot ranked
  8. MathNot ranked
Top decileTop quartileMid-fieldNot eligible

Benchmark ledger

Every result says who ran the test. For this model, Reflection AI ran 19 of 22. Coding opens by default. The tick marks the best source-verified result in the catalog; provisional leaders do not set the reference.

Overall rank
#132
of 214 on the leaderboard
Reported by Reflection AI
19
Shown with that label
Run by a benchmark
3
The benchmark’s own run
Independent runs
0
None yet
Coding8 rows
Coding benchmark results, best verified comparison, who ran each test, and whether it counts toward a rank
VulcanBench Frontier v464.7%

Best verified: Claude Fable 5.2 · 93.8%

Gap29.1 behindRun by VulcanBench Frontier v4Counts3% ref. weight
SWE-bench Pro66.1%

Best verified: Claude Fable 5.2 · 93%

Gap26.9 behindReported by Reflection AIShown only
DeepSWE43.3%

Best verified: Muse Wonder Full · 79.0%

Gap35.7 behindReported by Reflection AIShown only
SciCodeScientific Code Benchmark49.2%

Best verified: Sakana Fugu · 60.1%

Gap10.9 behindReported by Reflection AIShown only
SWE Multilingual78.3%

Best verified: Claude Fable 5.2 · 96.6%

Gap18.3 behindReported by Reflection AIShown only
Terminal-Bench 2.180.1%

Best verified: SWE-2 · 92.8%

Gap12.7 behindReported by Reflection AIShown only
SWE-Atlas Codebase QnA31.8%

Best verified: Ling 3.1 Flash · 64.5%

Gap32.7 behindReported by Reflection AIShown only
SWE-bench VerifiedSoftware Engineering Benchmark Verified81.6%

Best verified: Claude Opus 5 · 96%

Gap14.4 behindReported by Reflection AIShown only
Agentic7 rows
Agentic benchmark results, best verified comparison, who ran each test, and whether it counts toward a rank
Terminal-Bench 4.16.40%

Best verified: Claude Fable 5.2 · 65.80%

Gap59.4 behindRun by Terminal-Bench 4.1Counts9% ref. weight
HLE w/ toolsHumanity's Last Exam with tools51.1%

Best verified: Kimi K3.1 Pro · 70.0%

Gap18.9 behindReported by Reflection AICounts3% ref. weight
τ³-bench resultsτ³-Bench Tool-Agent-User Evaluation36.1%

Best verified: Mercury 2.5 · 96.0%

Gap59.9 behindReported by Reflection AICounts2% ref. weight
Terminal-Bench 2.180.1%

Best verified: SWE-2 · 92.8%

Gap12.7 behindReported by Reflection AIShown only
BrowseComp76.8%

Best verified: Atria Dawn Preview · 92.5%

Gap15.7 behindReported by Reflection AIShown only
MCP Atlas80.8%

Best verified: Muse Spark Open · 89.2%

Gap8.4 behindReported by Reflection AIShown only
DeepSearchQA80.5%

Best verified: Atria Dawn Preview · 96.0%

Gap15.5 behindReported by Reflection AIShown only
Reasoning2 rows
Reasoning benchmark results, best verified comparison, who ran each test, and whether it counts toward a rank
CritPtCritical Physics Tasks16.3%

Best verified: Claude Fable 5.2 · 35.4%

Gap19.1 behindReported by Reflection AICounts25% ref. weight
LongBench v267.5%

Best verified: Inkling 2 Small · 70.6%

Gap3.1 behindReported by Reflection AIShown only
Knowledge2 rows
Knowledge benchmark results, best verified comparison, who ran each test, and whether it counts toward a rank
HLE w/o toolsHumanity's Last Exam without tools36%

Best verified: Claude Fable 5.2 · 73.2%

Gap37.2 behindReported by Reflection AIShown only
GPQA-DGPQA Diamond90.6%

Best verified: GPT-6.2 Astra · 97.0%

Gap6.4 behindReported by Reflection AIShown only
Inst. Following1 row
Inst. Following benchmark results, best verified comparison, who ran each test, and whether it counts toward a rank
IFBenchInstruction Following Benchmark79.6%

Best verified: Flint · 89.2%

Gap9.6 behindReported by Reflection AIShown only
Math1 row
Math benchmark results, best verified comparison, who ran each test, and whether it counts toward a rank
AIME26AIME 202697.7%

Best verified: GLM-5.2 · 99.2%

Gap1.5 behindReported by Reflection AIShown only
External signals1 row
External signals benchmark results, best verified comparison, who ran each test, and whether it counts toward a rank
StealthMarkStealthMark Intelligence Benchmarks (overall score)31.9

Best verified: GPT-6 Astra · 74.6

Gap42.7 behindRun by StealthMarkShown only

Outlined bars are lab-reported. Filled bars are benchmark-run or independent.

Lineage

The sequence follows explicit supersedes links. Each score is estimated for that model; a relative can inform a sparse estimate but never sets a floor, so a newer release can score below an earlier one. Scores and prices remain blank when the corresponding public row or first-party rate is unavailable.

  1. Oct 16, 2026 · you are here

    Beam 1.0 501B-A23B

    Score 41.1 · Price not listed

Spec sheet

Each documented value carries its source. Missing fields stay visible as not sourced or not published, rather than disappearing from the page.

API model ID
Beam-1.0-501B-A23BReflection developer docs, Models
Maximum output
Not sourced yet
Knowledge cutoff
Not sourced yet
Parameters
Not sourced yet
Availability
Hugging Face · self-hosted inference · Reflection API public beta · partner endpointsReflection Beam 1.0 BF16 model card
Cloud regions
Not tracked yet
API capabilities
Tool calling, structured outputs, and batch support are not tracked yet
Prompt caching
Not documented in the pricing recordReflection launch post
Self-host
Beam 1.0 501B-A23B needs ~640GB VRAM (8× NVIDIA H100 (80GB)).
Rate limits
Not tracked yet

How to read this profile

The visual layer above carries the decisions. These notes preserve the model, ranking, coverage, and family context behind the numbers.

Beam 1.0 501B-A23B ranks #132 of 214 on the public leaderboard with a score of 41.15/100. It does not yet have enough sourced coverage for a verified position.

Beam 1.0 501B-A23B is a open weight model with a 1M context window. It uses an explicit reasoning mode, which can improve complex problem solving while adding latency and token use.

Open Apache 2.0 weights in BF16, FP8 and NVFP4 on Hugging Face for self-hosted inference (Reflection documents four NVLink-connected B300 GPUs for a single server), the Reflection API in public beta, and partner endpoints at launch. No per-token tariff is published for the first-party API.

Reflection released Beam 1.0 as open weights on October 16, 2026 under Apache 2.0: a text-only sparse mixture-of-experts model with 501.3B total and 23.4B active parameters, 52 layers, gated grouped-query and sliding-window attention, a 1M-token context window and a June 30, 2026 knowledge cutoff, published in BF16, FP8 and NVFP4 numerics with a DFlash2 speculative drafter. Benchmark values are the BF16 column of Reflection's model card, run at temperature 0.7 and top_p 0.99 with reasoning on; the FP8 and NVFP4 columns sit on their own rows. SWE-bench Pro v1 stays separate from v2-Hard, Terminal-Bench 2.1 is never mapped to 2.0, and the long-context, physics and Tau3 banking lanes carry the provider-run figures because no independent row exists yet. LiveCodeBench is reported without a version, SWE-bench Pro v2-Hard has no lane, AutomationBench and the Omniscience index are public-split-only figures, and the Iris safety rates are not benchmark lanes; those stay unmapped. The hosted API lists the model as Beam-1.0-501B-A23B in public beta with mandatory reasoning at five effort levels and a 128K maximum output.

Beam 1.0 501B-A23B sits in the Beam family with Beam 1.0 501B-A23B FP8, Beam 1.0 501B-A23B NVFP4. 22 of 923 tracked benchmark slots currently have displayable evidence. Missing categories stay blank.

Its strongest eligible category is Agentic at #74, while its lowest eligible position is Reasoning at #99. particularly useful for coding agents, browser research, and computer-use workflows.

Last updated November 19, 2026. Runtime fields remain blank until a sourced snapshot exists.

Deployment options

Self-host and provider-specific paths stay separate from benchmark evidence so operating constraints are visible before a score becomes the whole decision.

Where to run Beam 1.0 501B-A23B by API

3 hosts publish a per-token price for it. Nebius Token Factory has the lowest output rate, $1.00 per million tokens. Prices read from each host’s own pages on November 16, 2026.

Inference hosts for Beam 1.0 501B-A23B with input, output and cached-input prices per million tokens.
HostInput /1MOutput /1MCached /1MContextListed speed
Nebius Token Factory$0.25$1.00$0.051.05M611 tok/s
OpenRouter$0.30$1.00$0.041.05M—
Together AI$0.40$1.20$0.041.05M—
Compare inference hosts

Self-host vs API cost

Estimates at 50,000 req/day · 1000 tokens/req average.

Beam 1.0 501B-A23B
API / mo (Nebius Token Factory)$938
Self-host / mo$7,008
Break-evenNot reached at 70% load
Model the full break-even

Questions

How does Beam 1.0 501B-A23B perform overall in AI benchmarks?

Beam 1.0 501B-A23B ranks #132 out of 214 models on the public BenchAlign leaderboard, with a score of 41.15/100. Its evidence status is Estimated, and this profile shows 22 source-displayable benchmark rows. The label describes evidence depth, not a provider quality claim; inspect category rows before choosing a workload.

Is Beam 1.0 501B-A23B good for knowledge and understanding?

Beam 1.0 501B-A23B ranks #92 out of 179 eligible models for knowledge and understanding, with a public category score of 46.2/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.

Is Beam 1.0 501B-A23B good for coding and programming?

Beam 1.0 501B-A23B ranks #93 out of 160 eligible models for coding and programming, with a public category score of 32.5/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.

Is Beam 1.0 501B-A23B good for mathematics?

Beam 1.0 501B-A23B has source-displayable benchmark coverage for mathematics, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.

Is Beam 1.0 501B-A23B good for reasoning and logic?

Beam 1.0 501B-A23B ranks #99 out of 159 eligible models for reasoning and logic, with a public category score of 37.4/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.

Is Beam 1.0 501B-A23B good for agentic tool use and computer tasks?

Beam 1.0 501B-A23B ranks #74 out of 138 eligible models for agentic tool use and computer tasks, with a public category score of 40.8/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.

Is Beam 1.0 501B-A23B good for instruction following?

Beam 1.0 501B-A23B has source-displayable benchmark coverage for instruction following, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.

Is Beam 1.0 501B-A23B open source?

Beam 1.0 501B-A23B is an open-weight model from Reflection AI. Its weights can be downloaded for local or hosted deployment, subject to the published license. Open weight does not automatically mean open source: training data and training code may remain private, and commercial restrictions can still apply.

Which sibling models are related to Beam 1.0 501B-A23B?

Beam 1.0 501B-A23B belongs to the Beam family. Related tracked variants include Beam 1.0 501B-A23B FP8, Beam 1.0 501B-A23B NVFP4. A sibling link indicates shared lineage or a documented configuration relationship; it does not mean the variants have identical pricing, context limits, benchmark evidence, or deployment behavior. Compare before switching.

Does Beam 1.0 501B-A23B have full benchmark coverage on BenchLM?

No. Beam 1.0 501B-A23B currently has 41 source-displayable rows. The profile exposes published, non-generated evidence and leaves missing categories blank until an exact evaluation is available. Coverage describes how much was measured; it is not a penalty added to an individual benchmark result.

What is the context window size of Beam 1.0 501B-A23B?

Beam 1.0 501B-A23B has a documented context window of 1M. That figure is the maximum combined prompt and retained-conversation space reported for this exact model; it is not the maximum output length. The profile keeps output limits separate because providers often publish those limits independently.

Compare Beam 1.0 501B-A23B with every tracked model1032 comparisons