Beam 1.0 501B A23B — Hardware Requirements & GPU Compatibility
ChatReasoningCodeFunctions
Beam 1.0 is an open model from Reflection, a sparse mixture-of-experts language model with about 501 billion total parameters and 23 billion active per token. It takes text in and produces text out, and is aimed at software engineering, agentic tool use, and math and science reasoning. The company's main claim is token-efficient reasoning: it reaches comparable quality to leading open models while using around three times fewer reasoning tokens. The release includes BF16, FP8, and NVFP4 weights. Even at 4-bit it needs roughly 250 GB of memory, so it belongs on a multi-GPU server or a very large unified-memory machine. The context window is 1 million tokens, using gated grouped-query attention with sliding-window layers to keep long-context memory manageable. It is released under the Apache 2.0 license, so commercial use is unrestricted. The card says it works with agentic coding harnesses and MCP-based tools, and that it was developed mainly for English.
Beam 1.0 501B A23B runs on a 512 GB+ Mac or a multi-GPU server
Best local pick → Mac Studio (M3 Ultra, 512GB) · Q4_K_M · 304 GB · ~29 tok/s
Apache 2.0
Not set up for it?
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Beam 1.0 501B A23B
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
vllm serve reflection/Beam-1.0-501B-A23B --max-model-len 16384 --tensor-parallel-size 8Size the nodes from the table above
Download: Beam-1.0-501B-A23B-Q4_K_M.gguf300.76 GBunsloth
Other ways
each opens step-by-stepRun in cloud
Rent 5× H100 SXM · Q4_K_M
- Estimated hourly price
- ~$10.81/hr
- est., whole server
- GPUs
- 5 GPUs
- 400 GB of VRAM
How Much VRAM Does Beam 1.0 501B A23B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | Runs on | |
|---|---|---|---|---|
| 138.5 GB | 144 GB | 2-bit · i-quant extreme compression — Importance-weighted 2-bit, extreme compression — significant quality loss | Runs on GH200 Grace Hopper Superchip 144 GB | |
| 169.8 GB | 175.3 GB | 2-bit · i-quant medium — Importance-weighted 2-bit, medium | Runs on B200 192 GB | |
| 194.9 GB | 200.4 GB | 3-bit · i-quant — Importance-weighted 3-bit | Runs on Instinct MI325X 256 GB | |
| 213.7 GB | 219.2 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Runs on Instinct MI325X 256 GB | |
| 245.0 GB | 250.5 GB | 3-bit medium — 3-bit medium quantization | Runs on Instinct MI325X 256 GB | |
| 301.4 GB | 306.9 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Runs on 5× H100 SXM 80 GB | |
| 357.8 GB | 363.3 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Runs on 6× H100 SXM 80 GB | |
| 414.2 GB | 419.7 GB | 6-bit, very good — 6-bit quantization, very good quality | Runs on 6× H100 SXM 80 GB | |
| 501.9 GB | 507.4 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Runs on 8× H100 SXM 80 GB | |
| 1003.2 GB | 1008.7 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Runs on 8× H200 SXM 141 GB |
Selected Q4_K_M
4-bit medium quantization304.0GBat 16K context
- Weights
- 300.8 GB
- KV cache
- 2.9 GB
- Overhead
- 0.3 GB
Server-class: ~304 GB at 16K context · about 5× H100 SXM 80 GB
Which Hardware Can Run Beam 1.0 501B A23B?
Q4_K_M · 301.4 GBBeam 1.0 501B A23B (Q4_K_M) requires 301.4 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 392+ GB is recommended. Using the full 1049K context window can add up to 185.8 GB, bringing total usage to 487.2 GB. No consumer GPU has enough memory. 3 devices with unified memory can run Beam 1.0 501B A23B, including NVIDIA DGX H100, NVIDIA DGX A100 640GB, Mac Studio (M3 Ultra, 512GB).
192+ GB
Q4_K_M · best fit IQ2_M (172.4 GB)
6 GPUs · 1 Mac class · 2 device families
B200~228Simulate output on B200B300~228Simulate output on B300Instinct MI350X~83Simulate output on Instinct MI350XM3 Ultra256–512 GB~41Simulate output on Mac Studio (M3 Ultra, 256GB)8x H100 SXM5~302Simulate output on NVIDIA DGX H1008x A100 SXM4~277Simulate output on NVIDIA DGX A100 640GB
By VRAM tier · typical GPUs and Macs at Q4_K_M
Consumer GPUs (up to 32 GB) and Macs up to 128 GB: too small for this model (needs ≥ 141.1 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →
How good is Beam 1.0 501B A23B?
llmrun Score blends public benchmark boards into one number; 100 = the average open model.
Not enough data yet for an llmrun Score.
A Score needs results on enough independent boards. How the Composite Score is built
Benchmarks
This is the single board showing results for this model; badges show who ran the test.
- LMArena WebDev
LMArena · Oct 2026Arena
Arena1437.9
#22/52 open
Specifications
- Publisher
- reflection
- Parameters
- 501.3B
- Active parameters
- 23.4B per token
- Architecture
- BeamForCausalLM
- Context Length
- 1,048,576 tokens
- Vocabulary Size
- 151,936
- Release Date
- 2026-10-07
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Beam 1.0 501B A23B need?
Beam 1.0 501B A23B requires 301.4 GB of VRAM at Q4_K_M, or 1003.2 GB at BF16. Full 1049K context adds up to 185.8 GB (487.2 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 501.3B × 4.8 bits ÷ 8 = 300.8 GB
KV Cache + Overhead ≈ 0.6 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 186.4 GB (at full 1049K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 5090 run Beam 1.0 501B A23B?
No — Beam 1.0 501B A23B requires at least 138.5 GB at IQ2_XXS, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.
What's the best quantization for Beam 1.0 501B A23B?
For Beam 1.0 501B A23B, Q4_K_M (301.4 GB) offers the best balance of quality and VRAM usage. Q5_K_M (357.8 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 138.5 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Beam 1.0 501B A23B on a Mac?
Yes, but only at lower quantizations. The smallest Mac that can run Beam 1.0 501B A23B is Mac Pro M2 Ultra (192 GB) at IQ2_XXS; 4 of the 156 Macs we list can run it at some quantization. For Q4_K_M (301.4 GB) you need a Mac with more unified memory.
Can I run Beam 1.0 501B A23B locally?
Yes — Beam 1.0 501B A23B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
What's the download size of Beam 1.0 501B A23B?
At Q4_K_M, the download is about 300.76 GB. The full-precision BF16 version is 1002.54 GB. The smallest option (IQ2_XXS) is 137.85 GB.
Which GPUs can run Beam 1.0 501B A23B?
No single consumer GPU has enough VRAM to run Beam 1.0 501B A23B at Q4_K_M (301.4 GB). Multi-GPU or professional hardware is required.
Which devices can run Beam 1.0 501B A23B?
3 devices with unified memory can run Beam 1.0 501B A23B at Q4_K_M (301.4 GB), including Mac Studio (M3 Ultra, 512GB), NVIDIA DGX A100 640GB, NVIDIA DGX H100. Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.