Skip to content
Model Pareto

Beam 1.0 501B A23B

Reflection AIOpen weightshuggingface.co ↗
Compare with nearest rivalsCompare with…
Input
—
Output
—
Speed
~86tok/sest.
Context
1Mtokens

Rankings

Self-hostingEstimated

How we estimate →

What it would cost to run these open weights yourself on rented GPUs. No API sells this model, so this is its price on the chart.

Hardware
8× B200 180GB
FP8 weights
Throughput
~35,020 tok/s
many requests batched
Price
$0.16–1.46 /M tok
busy → light use
On your own machine
Mac Studio M5 Ultra 512GB (int4, ~72 tok/s single-stream)
single consumer GPU or Mac
Assumptions (5)
  • FP8 weights (501.3B params, 23.4B active per token) + 40% KV-cache headroom ≈ 702 GB
  • 8× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
  • ~35,020 output tok/s aggregate at batch 1002 (bandwidth-bound); MoE compute scales with active params
  • Blended 3:1 input:output; prefill ~363,462 tok/s
  • Low = 75% utilization at the low GPU price; high = 20% at the high price

Benchmark scores

~ italic, dashed = no published score yet, estimated from related benchmarks.

Coding

AA Coding Agent IndexEstimated
~#19 of 24 (estimated)~42.4
  1. #1Claude Opus 5.561.7
  2. #2Claude Opus 559.7
  3. #17Step 5 Preview43.1
  4. ~#19Beam 1.0 501B A23BEstimated42.4
  5. #19Kimi K341.4
No published score — estimated from related benchmarks

Writing & chat

MMMLUEstimated
~#2 of 5 (estimated)~90.8%
  1. #1Gemini 3.1 Pro92.6%
  2. ~#2Beam 1.0 501B A23BEstimated90.8%
  3. #2Gemma 4 31B88.4%
No published score — estimated from related benchmarks
LMArena Text, Non-English (Elo)Estimated
~#18 of 57 (estimated)~1454 Elo
  1. #1Gemini 4 ArgonFew votes1512
  2. #2Claude Opus 5.51497
  3. #17GPT-5.6 Terra1456
  4. ~#18Beam 1.0 501B A23BEstimated1454
  5. #18DeepSeek V4 Pro (0813)1448
No published score — estimated from related benchmarks
LMArena Text (Elo)Estimated
~#18 of 57 (estimated)~1465 Elo
  1. #1Gemini 4 Argon1525
  2. #2Claude Opus 5.51507
  3. #17GPT-5.6 Terra1466
  4. ~#18Beam 1.0 501B A23BEstimated1465
  5. #18DeepSeek V4 Pro (0813)1463
No published score — estimated from related benchmarks
EQ-Bench Creative Writing v3 (Elo)Estimated
~#9 of 26 (estimated)~1953 Elo
  1. #1GPT-6 Astra2173
  2. #2Claude Fable 5.12162
  3. #8Grok 4.72007
  4. ~#9Beam 1.0 501B A23BEstimated1953
  5. #9Muse Spark 1.31906
No published score — estimated from related benchmarks

Hard reasoning

Agents

τ²-bench (Telecom)Estimated
~#3 of 120 (estimated)~97.6%
  1. #1Step 3.7 Flash98.5%
  2. #2DeepSeek-V4-Pro-0813-nvfp4-DSpark98.3%
  3. ~#3Beam 1.0 501B A23BEstimated97.6%
  4. #3Gemini 3.1 Pro95.6%
No published score — estimated from related benchmarks
OSWorld-VerifiedEstimated
~#12 of 15 (estimated)~65.9%
  1. #1Kimi K384.8%
  2. #2Qwen3.8 27B84.3%
  3. #11Muse Glimmer65.9%
  4. ~#12Beam 1.0 501B A23BEstimated65.9%
  5. #12Nemotron 3.5 Super VL NVFP452.3%
No published score — estimated from related benchmarks
OSWorld 2.0Estimated
~last of 5 (estimated)below measured range (<47.5%)
  1. #1GLM-5.3 Flash59.1%
  2. #2Kimi K358.3%
  3. #4GPT-5.547.5%
  4. ~#5Beam 1.0 501B A23BEstimated<47.5%
No published score — estimated from related benchmarks
GDPval-AA (Elo)Estimated
~#18 of 91 (estimated)~1576 Elo
  1. #1Claude Opus 5.51866
  2. #2Muse Wonder Full1857
  3. #17Step 5 Preview1584
  4. ~#18Beam 1.0 501B A23BEstimated1576
  5. #18GPT-6 Astra1574
No published score — estimated from related benchmarks
AutomationBench-AAEstimated
~#26 of 84 (estimated)~50.4%
  1. #1Muse Wonder Full71.3%
  2. #2Claude Opus 5.569.5%
  3. #25Step 5 Preview51.0%
  4. ~#26Beam 1.0 501B A23BEstimated50.4%
  5. #26Qwen3.8 27B48.2%
No published score — estimated from related benchmarks
τ-Bench Banking (AA)Estimated
~#18 of 91 (estimated)~36.7%
  1. #1Muse Spark 1.350.5%
  2. #2GLM-5.350.3%
  3. #17Claude Sonnet 537.3%
  4. ~#18Beam 1.0 501B A23BEstimated36.7%
  5. #18Motif 335.3%
No published score — estimated from related benchmarks