ModelBench
Reflection logo
Models/Reflection

Reflection: Beam 1.0 501B A23B

Sparse mixture-of-experts reasoning model with 23B active parameters for coding and agentic workflows

Source-linked beam Open weights Apache-2.0 Released 2026-10-05
API record Report
InputT
OutputT
Input price$0.24/M
Output price$0.98/M
Context1.04858M
Max output131.072K
Providers9
Inference availability

Providers

Provider-specific identifiers, limits, and listed prices per million tokens. Every row links back to the provider's own documentation.

Report a price
Providers offering Beam 1.0 501B A23B
ProviderProvider model IDContextMax outputInputOutputCache readCapabilitiesDocs
Reflection Beam-1.0-501B-A23B 1.04858M 131.072K — — — ReasoningToolsJSON Docs ↗
Cloudflare AI Gateway reflection-ai/beam-1.0-501b-a23b 262.144K 131.072K — — — ReasoningToolsJSON Docs ↗
Ollama Cloud beam 1.04858M 131.072K — — — ReasoningTools Docs ↗
Kilo Gateway reflection/beam-1.0-501b-a23b 1.04858M 131.072K $0.24 $0.98 $0.035 ReasoningToolsJSON Docs ↗
OpenRouter reflection/beam-1.0-501b-a23b 1.04858M 131.072K $0.25 $1 $0.05 ReasoningToolsJSON Docs ↗
Nebius Token Factory reflection/beam 1.04858M 131.072K $0.25 $1 $0.05 ReasoningToolsJSON Docs ↗
CoreWeave reflection/Beam-1.0-501B-A23B 1.04858M 1.04858M $0.3 $1 $0.04 ReasoningToolsJSON Docs ↗
Together AI reflection/Beam-1.0-501B-A23B 1.04858M 131.072K $0.4 $1.2 $0.04 ReasoningToolsJSON Docs ↗
NanoGPT reflection/beam-1.0 1.04858M 131.072K $0.5 $1.5 $0.05 ReasoningToolsJSON Docs ↗

Capability badges appear only where the provider catalog explicitly lists support. A blank cell means the source is silent, not that the feature is absent.

Listed rates

Price across providers

Input price per million tokens as published by each provider. Bars are drawn from listed rates only — no traffic weighting, since the catalog observes no requests.

Lowest input $0.24/M

Across 6 priced providers

Median input $0.275/M

Midpoint of listed rates

Highest input $0.5/M

2.1× the lowest listed rate

Output range $0.98 – $1.5

Per million output tokens

Kilo Gateway $0.24/MLowest
OpenRouter $0.25/M
Nebius Token Factory $0.25/M
CoreWeave $0.3/M
Together AI $0.4/M
NanoGPT $0.5/M
Cost calculator

Estimate a workload

$0.00
Excludes taxes, non-token charges, and tiered discounts.

Context limits also differ by provider, from 262.144K to 1.04858M tokens. Compare the provider table above before choosing on price alone.

Specification

Capabilities

Recorded from the source catalog and provider listings.

✓ Reasoning Yes
✓ Tool calling Yes
× Structured output No
× Attachments No
× Vision input No
✓ Open weights Yes
Creator
Reflection
Model family
beam
Knowledge cutoff
2026-06-30
License
Apache-2.0
Release date
2026-10-05
Model ID
reflection/beam-1.0-501b-a23b

Weights: Hugging Face ↗ · Hugging Face (FP8) ↗ · Hugging Face (NVFP4) ↗

Published evaluations

Benchmarks

Every result stays attached to its source, version, metric and harness. Scores from different versions are never merged, and the profile below plots each benchmark against its own population rather than on a shared scale.

Benchmark registry
Benchmark profile
Design Arena: w…Design Arena: website Design Arena: g…Design Arena: gamedev Humanity's Last…Humanity's Last Exam DeepSWEDeepSWE 1.1 GPQA DiamondGPQA Diamond IFBenchIFBench MCP AtlasMCP Atlas SciCodeSciCode This model — Design Arena: website: 1242 elo (59.8th percentile) This model — Design Arena: gamedev: 1210 elo (55.8th percentile) This model — Humanity's Last Exam: 51.1 score (45.8th percentile) This model — DeepSWE 1.1: 43.3 score (18.8th percentile) This model — GPQA Diamond: 90.6 score (43.8th percentile) This model — IFBench: 79.6 score (30.0th percentile) This model — MCP Atlas: 80.8 score (70.0th percentile) This model — SciCode: 49.2 score (10.0th percentile)

Each axis is this model's percentile among the 8 benchmarks it has published results for, measured against every other model with a score on that same benchmark. Percentiles are used because benchmarks do not share a scale — a 60 on one is not a 60 on another. Hover any point for the raw score.

Design Arena: website elo Source ↗
Design Arena: website #81 of 200
Design Arena: gamedev elo Source ↗
Design Arena: gamedev #85 of 191
Humanity's Last Exam score · 2026-10-16 Source ↗
Humanity's Last Exam #7 of 12
DeepSWE — 1.1 score · 2026-10-16 Source ↗
DeepSWE #7 of 8
GPQA Diamond score · 2026-10-16 Source ↗
GPQA Diamond #5 of 8
IFBench score · 2026-10-16 Source ↗
IFBench #4 of 5
MCP Atlas score · 2026-10-16 Source ↗
MCP Atlas #2 of 5
SciCode score · 2026-10-16 Source ↗
SciCode #5 of 5
Terminal-Bench — 2.1 score · 2026-10-16 Source ↗
Terminal-Bench #3 of 4
LiveCodeBench score · 2026-10-16 Source ↗
LiveCodeBench #3 of 3
τ3 Banking score · 2026-10-16 Source ↗
τ3 Banking #1 of 3
AA-LCR score · 2026-10-16 Source ↗
78.2 2 models scored — too few for a distribution
AutomationBench score · 2026-10-16 Source ↗
39.3 2 models scored — too few for a distribution
LongBench — 2 score · 2026-10-16 Source ↗
67.5 2 models scored — too few for a distribution
AA-Omniscience index · 2026-10-16 Source ↗
10.8 1 model scored — too few for a distribution
AIME 2026 score · 2026-10-16 Source ↗
97.7 1 model scored — too few for a distribution
BrowseComp score · 2026-10-16 Source ↗
76.8 1 model scored — too few for a distribution
CritPt score · 2026-10-16 Source ↗
16.3 1 model scored — too few for a distribution
DeepSearchQA score · 2026-10-16 Source ↗
80.5 1 model scored — too few for a distribution
SWE-Atlas Codebase QnA score · 2026-10-16 Source ↗
31.8 1 model scored — too few for a distribution
SWE-Bench Multilingual score · 2026-10-16 Source ↗
78.3 1 model scored — too few for a distribution
SWE-Bench Pro — 1 score · 2026-10-16 Source ↗
66.1 1 model scored — too few for a distribution
SWE-Bench Pro — 2 score · 2026-10-16 Source ↗
76.5 1 model scored — too few for a distribution
SWE-Bench Verified score · 2026-10-16 Source ↗
81.6 1 model scored — too few for a distribution

Each strip shows every published score for that benchmark, with this model marked. The lighter shape behind the ticks is the density of results, and the dashed line is the median.

Cost against capability

Price and performance

Listed input price plotted against Design Arena: website, the benchmark with the widest published coverage that this model appears in.

0 721 1443 $0.1 $1 $10 Amazon: Nova Premier 1.0 — $2.5/M, score 835.0 Nova Pro 1.0 — $0.799/M, score 796.0 Claude Fable 5 — $10/M, score 1294.0 Anthropic: Claude Fable 5.1 — $10/M, score 1311.0 Anthropic: Claude Fable 5.1 (batch) — $5/M, score 1311.0 Anthropic: Claude Fable 5 (batch) — $5/M, score 1294.0 Anthropic: Claude Haiku 4.5 — $1/M, score 1122.0 Anthropic: Claude Haiku 4.5 (batch) — $0.5/M, score 1122.0 Anthropic: Claude Haiku 5.5 — $0.1/M, score 1261.0 Anthropic: Claude Haiku 5.5 (batch) — $0.05/M, score 1261.0 Anthropic: Claude Opus 4 — $15/M, score 1176.0 Anthropic: Claude Opus 4.1 — $15/M, score 1175.0 Anthropic: Claude Opus 4.1 (batch) — $7.5/M, score 1175.0 Anthropic: Claude Opus 4.5 — $5/M, score 1245.0 Anthropic: Claude Opus 4.5 (batch) — $2.5/M, score 1245.0 Anthropic: Claude Opus 4.6 — $5/M, score 1289.0 Anthropic: Claude Opus 4.6 (batch) — $2.5/M, score 1289.0 Anthropic: Claude Opus 4.7 — $5/M, score 1307.0 Anthropic: Claude Opus 4.7 (batch) — $2.5/M, score 1307.0 Anthropic: Claude Opus 4.8 — $5/M, score 1257.0 Anthropic: Claude Opus 4.8 (batch) — $2.5/M, score 1257.0 Claude Opus 5 — $5/M, score 1308.0 Anthropic: Claude Opus 5.5 — $4/M, score 1329.0 Anthropic: Claude Opus 5.5 (batch) — $2/M, score 1329.0 Anthropic: Claude Opus 5 (batch) — $2.5/M, score 1308.0 Anthropic: Claude Sonnet 4 — $3/M, score 1147.0 Anthropic: Claude Sonnet 4.5 — $3/M, score 1188.0 Anthropic: Claude Sonnet 4.5 (batch) — $1.5/M, score 1188.0 Anthropic: Claude Sonnet 4.6 — $3/M, score 1283.0 Anthropic: Claude Sonnet 4.6 (batch) — $1.5/M, score 1283.0 Claude Sonnet 5 — $2/M, score 1273.0 Anthropic: Claude Sonnet 5.5 — $2/M, score 1325.0 Anthropic: Claude Sonnet 5.5 (batch) — $1/M, score 1325.0 Anthropic: Claude Sonnet 5 (batch) — $1/M, score 1273.0 Trinity Large Thinking — $0.25/M, score 1137.0 DeepSeek Chat — $0.15/M, score 1118.0 DeepSeek V3.1 — $0.19/M, score 1122.0 DeepSeek R1 0528 — $0.2/M, score 1149.0 DeepSeek: DeepSeek V3.1 Terminus — $0.27/M, score 1185.0 DeepSeek V3.2 — $0.209/M, score 1173.0 DeepSeek: DeepSeek V3.2 Exp — $0.27/M, score 1177.0 DeepSeek V4 Flash — $0.15/M, score 1206.0 DeepSeek V4 Flash 0731 — $0.035/M, score 1236.0 DeepSeek: DeepSeek V4 Flash 0731 (batch) — $0.11/M, score 1250.0 DeepSeek V4 Pro — $0.66/M, score 1234.0 DeepSeek V4.1 Flash 1017 — $0.15/M, score 1309.0 Gemini 2.5 Flash — $0.3/M, score 1113.0 Google: Gemini 2.5 Flash (batch) — $0.15/M, score 1115.0 Gemini 2.5 Pro — $1.25/M, score 1165.0 Google: Gemini 2.5 Pro (batch) — $0.625/M, score 1167.0 Gemini 3 Flash Preview — $0.5/M, score 1196.0 Google: Gemini 3 Flash Preview (batch) — $0.25/M, score 1196.0 Gemini 3.1 Flash Lite Preview — $0.25/M, score 1081.0 Gemini 3.1 Pro Preview — $2/M, score 1251.0 Google: Gemini 3.1 Pro Preview (batch) — $1/M, score 1251.0 Gemini 3.5 Flash — $1.5/M, score 1266.0 Google: Gemini 3.5 Flash (batch) — $0.75/M, score 1266.0 Gemini 3.6 Flash — $0.75/M, score 1299.0 Google: Gemini 3.6 Flash (batch) — $0.375/M, score 1299.0 Gemini 3.7 Flash — $0.75/M, score 1302.0 Google: Gemini 3.7 Flash (batch) — $0.375/M, score 1302.0 Gemini 3.8 Flash — $0.75/M, score 1299.0 Google: Gemini 3.8 Flash (batch) — $0.375/M, score 1299.0 Gemini 4 Argon — $2/M, score 1314.0 Google: Gemini 4 Argon (batch) — $1/M, score 1314.0 Mercury 2 — $0.25/M, score 1035.0 Mercury 3 — $0.1/M, score 1221.0 Meta: Llama 4 Maverick — $0.188/M, score 869.0 Meta: Llama 4 Scout — $0.1/M, score 750.0 Muse Glimmer 30B — $0.2/M, score 1147.0 Muse Spark 1.1 — $1.25/M, score 1269.0 Muse Spark 1.2 — $1.25/M, score 1311.0 Muse Spark 1.3 — $1.25/M, score 1355.0 Muse Spark Open — $1.25/M, score 1320.0 Meta: Muse Wonder — $4/M, score 1361.0 MiniMax: MiniMax M2 — $0.255/M, score 1141.0 MiniMax: MiniMax M2.1 — $0.3/M, score 1201.0 MiniMax: MiniMax M2.5 — $0.3/M, score 1222.0 MiniMax: MiniMax M2.7 — $0.3/M, score 1241.0 MiniMax: MiniMax M3 — $0.3/M, score 1255.0 MiniMax: MiniMax M3 (batch) — $0.3/M, score 1271.0 Mistral: Codestral 2508 — $0.3/M, score 1011.0 Mistral: Codestral 2508 (batch) — $0.15/M, score 1011.0 Mistral: Ministral 3 14B 2512 — $0.2/M, score 1081.0 Mistral: Ministral 3 3B 2512 — $0.1/M, score 1026.0 Mistral: Ministral 3 8B 2512 — $0.15/M, score 1063.0 Mistral: Ministral 3 8B 2512 (batch) — $0.075/M, score 1063.0 Mistral: Mistral Large 3 2512 — $0.5/M, score 1162.0 Mistral: Mistral Large 3 2512 (batch) — $0.25/M, score 1162.0 Mistral: Mistral Large 4 — $1.36/M, score 1257.0 Mistral: Mistral Medium 3 — $0.4/M, score 1077.0 Mistral: Mistral Medium 3.1 — $0.4/M, score 1131.0 Mistral: Mistral Medium 3.1 (batch) — $0.2/M, score 1131.0 Mistral: Mistral Small 3.2 24B — $0.075/M, score 894.0 Kimi K2 0711 — $0.4/M, score 1050.0 Kimi K2 Instruct 0905 — $0.6/M, score 1107.0 Kimi K2 Thinking — $0.4/M, score 1113.0 Kimi K2.5 — $0.375/M, score 1248.0 Kimi K2.6 — $0.95/M, score 1267.0 Kimi K2.7 Code — $0.95/M, score 1262.0 MoonshotAI: Kimi K2.7 Code (batch) — $0.475/M, score 1284.0 Kimi K3 — $3/M, score 1334.0 Kimi K3.1 Pro — $2/M, score 1328.0 MoonshotAI: Kimi K3 (batch) — $2.28/M, score 1334.0 Nex-N2-Pro — $0.5/M, score 1237.0 Nemotron 3 Ultra 550B A55B — $0.1/M, score 1144.0 NVIDIA: Nemotron 3 Ultra (batch) — $0.6/M, score 1145.0 NVIDIA: Nemotron 3.5 Super VL — $0.15/M, score 1172.0 GPT-4.1 — $2/M, score 1037.0 GPT-4.1 mini — $0.4/M, score 997.0 OpenAI: GPT-4.1 Mini (batch) — $0.2/M, score 997.0 GPT-4.1 nano — $0.1/M, score 971.0 OpenAI: GPT-4.1 Nano (batch) — $0.05/M, score 971.0 OpenAI: GPT-4.1 (batch) — $1/M, score 1037.0 GPT-4o — $2.5/M, score 831.0 OpenAI: GPT-4o (batch) — $1.25/M, score 831.0 GPT-5 — $1.25/M, score 1184.0 GPT-5 Mini — $0.25/M, score 1125.0 OpenAI: GPT-5 Mini (batch) — $0.125/M, score 1125.0 GPT-5 Nano — $0.05/M, score 1101.0 OpenAI: GPT-5 Nano (batch) — $0.025/M, score 1101.0 GPT-5.1 — $1.25/M, score 1185.0 GPT-5.1 Codex — $1.07/M, score 1160.0 GPT-5.1 Codex mini — $0.22/M, score 1109.0 OpenAI: GPT-5.1 (batch) — $0.625/M, score 1185.0 GPT-5.2 — $1.75/M, score 1193.0 OpenAI: GPT-5.2 (batch) — $0.875/M, score 1193.0 GPT-5.3 Codex — $1.75/M, score 1161.0 GPT-5.4 — $2.5/M, score 1218.0 OpenAI: GPT-5.4 (batch) — $1.25/M, score 1218.0 GPT-5.5 — $5/M, score 1255.0 OpenAI: GPT-5.5 (batch) — $2.5/M, score 1255.0 OpenAI: GPT-5 (batch) — $0.625/M, score 1184.0 GPT-6.1 Sol — $2/M, score 1320.0 OpenAI: GPT-6.1 Sol (batch) — $1/M, score 1320.0 GPT OSS 120B — $0.03/M, score 966.0 OpenAI: gpt-oss-120b (batch) — $0.03/M, score 966.0 GPT OSS 20B — $0.018/M, score 852.0 OpenAI: gpt-oss-20b (batch) — $0.024/M, score 852.0 o3 — $2/M, score 1036.0 OpenAI: o3 (batch) — $1/M, score 1036.0 o4-mini — $1.1/M, score 985.0 OpenAI: o4 Mini (batch) — $0.55/M, score 985.0 Qwen: Qwen3 235B A22B — $0.455/M, score 1035.0 Qwen: Qwen3 235B A22B Instruct 2507 — $0.09/M, score 1056.0 Qwen: Qwen3 235B A22B Thinking 2507 — $0.45/M, score 1051.0 Qwen: Qwen3 30B A3B — $0.12/M, score 953.0 Qwen: Qwen3 30B A3B Thinking 2507 — $0.2/M, score 935.0 Qwen: Qwen3 Coder 480B A35B — $0.3/M, score 1157.0 Qwen: Qwen3 Coder 30B A3B Instruct — $0.07/M, score 1087.0 Qwen: Qwen3 Max — $0.78/M, score 1123.0 Qwen: Qwen3.5 397B A17B — $0.5/M, score 1189.0 Qwen: Qwen3.5 Plus 2026-02-15 — $0.26/M, score 1187.0 Qwen: Qwen3.6 Plus — $0.325/M, score 1241.0 Qwen: Qwen3.7 Max — $1.475/M, score 1273.0 Qwen: Qwen3.7 Plus — $0.32/M, score 1268.0 Qwen: Qwen3.7 Plus (batch) — $0.23/M, score 1268.0 Qwen: Qwen3.8 Max — $2/M, score 1295.0 Beam 1.0 501B A23B — $0.24/M, score 1242.0 Step 3.7 Flash — $0.202/M, score 1195.0 Hy3 — $0.066/M, score 1180.0 Inkling — $1/M, score 1223.0 Thinking Machines: Inkling (batch) — $1/M, score 1229.0 Solar Mini 4 — $0.1/M, score 1153.0 Solar Pro 4 — $0.3/M, score 1176.0 SpaceXAI: Grok 4.20 — $1.25/M, score 1229.0 SpaceXAI: Grok 4.3 — $1.25/M, score 1194.0 SpaceXAI: Grok 4.3 (batch) — $1/M, score 1194.0 SpaceXAI: Grok 4.5 — $2/M, score 1279.0 SpaceXAI: Grok 4.6 — $2/M, score 1287.0 SpaceXAI: Grok 4.7 — $2/M, score 1252.0 MiMo-V2.5 — $0.14/M, score 1264.0 MiMo-V2.5-Pro — $0.435/M, score 1265.0 MiMo-V2.6-Pro — $0.435/M, score 1306.0 Z.ai: GLM 4.5 — $0.6/M, score 1168.0 Z.ai: GLM 4.5 Air — $0.13/M, score 1145.0 Z.ai: GLM 4.6 — $0.55/M, score 1174.0 Z.ai: GLM 4.7 — $0.54/M, score 1224.0 Z.ai: GLM 4.7 Flash — $0.061/M, score 1194.0 Z.ai: GLM 5 — $1/M, score 1247.0 Z.ai: GLM 5 Turbo — $1.2/M, score 1260.0 Z.ai: GLM 5.1 — $1.4/M, score 1290.0 Z.ai: GLM 5.2 — $0.034/M, score 1283.0 Z.ai: GLM 5.2 (batch) — $0.7/M, score 1302.0 Z.ai: GLM 5.3 — $0.079/M, score 1292.0 Z.ai: GLM 5.3 Flash — $0.15/M, score 1270.0 Z.ai: GLM 5.3 Flash (batch) — $0.05/M, score 1270.0 Z.ai: GLM 5.3 (batch) — $0.45/M, score 1292.0 Z.ai: GLM 5V Turbo — $1.2/M, score 1228.0 Beam 1.0 501B A23B Input price per million tokens (log scale) Elo

The stepped line is the efficient frontier: at each price, the best score available for that money or less. A model sitting on it is not being beaten by anything cheaper. Price is log-scaled because listed rates span four orders of magnitude. Only models with both a listed price and a score on this benchmark can appear.

Catalog activity

Change log

Field-level changes detected between successful source imports.

Full change log
Price Completion1.0 → 0.98
Price Prompt0.25 → 0.24
Price Completion0.98 → 1.0
Price Prompt0.24 → 0.25
Price Completion1.0 → 0.98
Price Prompt0.25 → 0.24
Price Completion0.98 → 1.0
Price Prompt0.24 → 0.25
Price Completion1.0 → 0.98
Price Prompt0.25 → 0.24
Provenance

Sources & verification

Every figure on this page traces back to one of these records.

Methodology
Last verified
Oct 24, 2026
Status
Source-linked
Confidence
Medium
Catalog source
Models.dev
Models.devcatalog ↗Reflection documentationprovider documentation ↗Cloudflare AI Gateway documentationprovider documentation ↗Ollama Cloud documentationprovider documentation ↗Kilo Gateway documentationprovider documentation ↗OpenRouter documentationprovider documentation ↗Nebius Token Factory documentationprovider documentation ↗CoreWeave documentationprovider documentation ↗Together AI documentationprovider documentation ↗NanoGPT documentationprovider documentation ↗τ3 Bankingoriginal repository ↗Design Arena: websiteindependent benchmark ↗
Public API

Use this record

Fetch the complete source-linked model record. No key, no account, no rate-limited tier.

API documentation
Endpoint
GET https://modelbench.lol/api/v1/models/reflection/beam-1.0-501b-a23b
curl
curl "https://modelbench.lol/api/v1/models/reflection/beam-1.0-501b-a23b"
Common questions

Frequently asked questions

Answered directly from the stored record — nothing here is generated beyond the catalog's own fields.

What is Beam 1.0 501B A23B?

Sparse mixture-of-experts reasoning model with 23B active parameters for coding and agentic workflows. It is published by Reflection and catalogued here from Models.dev.

How much does Beam 1.0 501B A23B cost?

Listed input pricing starts at $0.24 per million tokens from Kilo Gateway, rising to $0.5 across 6 listed providers.

What is the context length of Beam 1.0 501B A23B?

Beam 1.0 501B A23B accepts up to 1.04858M tokens of context and returns up to 131.072K output tokens.

Does Beam 1.0 501B A23B support tool calling and structured output?

Provider catalogs list support for tool calling, and reasoning.

Which providers serve Beam 1.0 501B A23B?

9 providers list this model: Reflection, Cloudflare AI Gateway, Ollama Cloud, Kilo Gateway, OpenRouter, Nebius Token Factory and 3 more.

Are the weights for Beam 1.0 501B A23B open?

Yes. The weights are published and downloadable from Hugging Face, Hugging Face (FP8) under the Apache-2.0 license.

When was Beam 1.0 501B A23B released?

The catalog records a release date of 2026-10-05, last verified Oct 24, 2026.