Step 3.5 Flash REAP 121B A11B — Hardware Requirements & GPU Compatibility
Chat
Step 3.5 Flash REAP 121B A11B is a 121.0B-parameter open language model from Cerebras in the Step 3 family. It supports a context window of up to 262,144 tokens. At Q4_K_M it needs about 79.84 GB of VRAM — see which GPUs and Macs can run it below.
Step 3.5 Flash REAP 121B A11B runs on a 96 GB+ Mac or a workstation GPU
Best local pick → Mac Studio (M3 Ultra, 96GB) · Q4_K_M · 85.3 GB · ~44 tok/s
Macs from 96 GB (Q4_K_M · ~44 tok/s)
Other
Not set up for it?
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Step 3.5 Flash REAP 121B A11B
One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.
Recommended
ollama run hf.co/mradermacher/Step-3.5-Flash-REAP-121B-A11B-i1-GGUF:Q4_K_MQ4_K_M runs on one RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB · 16K context
Download: Step-3.5-Flash-REAP-121B-A11B.i1-Q4_K_M.gguf72.82 GBmradermacher
Ways to runOther ways
each opens step-by-stepRun in cloud
Runs on H200 (141 GB) (55 GB headroom) · Q4_K_M
- Generation speed
- ~200 tok/s
- generation speed
- Cost per 1M output tokens
- $4.99
- per 1M output tokens
How Much VRAM Does Step 3.5 Flash REAP 121B A11B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | Runs on | |
|---|---|---|---|---|
| 56.5 GB | 68.3 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Runs on Instinct MI210 64 GB | |
| 64.9 GB | 76.6 GB | 3-bit medium — 3-bit medium quantization | Runs on RTX PRO 5000 Blackwell 72 GB | |
| 79.8 GB | 91.6 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Runs on RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB | |
| 94.8 GB | 106.6 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Runs on Instinct MI250X 128 GB | |
| 109.8 GB | 121.5 GB | 6-bit, very good — 6-bit quantization, very good quality | Runs on Instinct MI250X 128 GB | |
| 133.1 GB | 144.8 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Runs on H200 NVL 141 GB | |
| 266.1 GB | 277.9 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Runs on Instinct MI350X 288 GB |
Selected Q4_K_M
4-bit medium quantization85.3GBat 16K context
- Weights
- 78.8 GB
- KV cache
- 6.3 GB
- Overhead
- 0.3 GB
Server-class: ~85.3 GB at 16K context · one RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB
Or .
Which Hardware Can Run Step 3.5 Flash REAP 121B A11B?
Q4_K_M · 79.8 GBStep 3.5 Flash REAP 121B A11B (Q4_K_M) requires 79.8 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 104+ GB is recommended. No consumer GPU has enough memory. 30 devices with unified memory can run Step 3.5 Flash REAP 121B A11B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio (M3 Ultra, 96GB).
64–80 GB
Q4_K_M · best fit Q2_K (62.0 GB)
5 GPUs · 3 Mac classes · 1 device family
H100 SXM~198Simulate output on H100 SXMA100 80GB SXM~149Simulate output on A100 80GB SXMH100 PCIe~149Simulate output on H100 PCIeM2 Ultra*64 GB~53Simulate output on Mac Pro M2 Ultra (64 GB)M5 Max64 GB~45Simulate output on Mac Studio M5 Max (64 GB)Orin~23Simulate output on NVIDIA Jetson AGX Orin 64GB96 GB
fits · 10.7 GB free · up to 32K ctx
3 GPUs · 2 Mac classes
RTX PRO 6000 Blackwell Max-Q Workstation Edition~108Simulate output on RTX PRO 6000 Blackwell Max-Q Workstation EditionRTX PRO 6000 Blackwell Workstation Edition~108Simulate output on RTX PRO 6000 Blackwell Workstation EditionRTX PRO 6000 Blackwell Server Edition~99Simulate output on RTX PRO 6000 Blackwell Server EditionM3 Ultra*96 GB~44Simulate output on Mac Studio (M3 Ultra, 96GB)M3 Max*96 GB~22Simulate output on MacBook Pro 14" M3 Max (96 GB)128–144 GB
fits · 42.7 GB free · up to 64K ctx
4 GPUs · 2 Mac classes · 4 device families
GH200 Grace Hopper Superchip~202Simulate output on GH200 Grace Hopper SuperchipH200 NVL~200Simulate output on H200 NVLH200 SXM~200Simulate output on H200 SXMM2 Ultra*128 GB~43Simulate output on Mac Pro M2 Ultra (128 GB)M5 Max128 GB~37Simulate output on Mac Studio M5 Max (128 GB)GB10~21Simulate output on NVIDIA DGX SparkJetson T5000 (Thor)~21Simulate output on NVIDIA Jetson AGX Thor Developer Kit
By VRAM tier · typical GPUs and Macs at Q4_K_M
Consumer GPUs (up to 32 GB) and Macs up to 64 GB: too small for this model (needs ≥ 62 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →
Alternatives to Step 3.5 Flash REAP 121B A11B
Step 3 family · 6 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| GPT OSS 120BOpenAI | 97.5 | 70.8 GB | —needs > 24 GB | same size |
| Ornith 1.5 122Bornith-ai | — | 75.5 GB | —needs > 24 GB | same size |
| NVIDIA Nemotron 3.5 Super VL 120B A12B BF16NVIDIA | — | 87.7 GB | —needs > 24 GB | same size |
| Ling 3.0 Flash VLInclusion AI | — | 76 GB | —needs > 24 GB | same size |
| Leanstral 1.5 119B A6BMistral AI | — | 84 GB | —needs > 24 GB | same size |
| Laguna S 2.1poolside | — | 72.5 GB | —needs > 24 GB | same size |
Specifications
- Publisher
- Cerebras
- Family
- Step 3
- Parameters
- 121.0B
- Active parameters
- 12B per token
- Architecture
- Step3p5ForCausalLM
- Context Length
- 262,144 tokens
- Vocabulary Size
- 128,896
- Release Date
- 2026-02-25
- License
- Other
Frequently Asked Questions
How much VRAM does Step 3.5 Flash REAP 121B A11B need?
Step 3.5 Flash REAP 121B A11B requires 79.8 GB of VRAM at Q4_K_M, or 266.1 GB at BF16.
VRAM = Weights + KV Cache + Overhead
Weights = 121.0B × 4.8 bits ÷ 8 = 72.6 GB
KV Cache + Overhead ≈ 7.2 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.
VRAM usage by quantization
Can NVIDIA GeForce RTX 5090 run Step 3.5 Flash REAP 121B A11B?
No — Step 3.5 Flash REAP 121B A11B requires at least 56.5 GB at Q2_K, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.
What's the best quantization for Step 3.5 Flash REAP 121B A11B?
For Step 3.5 Flash REAP 121B A11B, Q4_K_M (79.8 GB) offers the best balance of quality and VRAM usage. Q5_K_M (94.8 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 56.5 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Step 3.5 Flash REAP 121B A11B on a Mac?
Yes — Mac Studio (M3 Ultra, 96GB) and 20 other Macs can run Step 3.5 Flash REAP 121B A11B. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 79.8 GB of usable unified memory (RAM minus macOS overhead).
Can I run Step 3.5 Flash REAP 121B A11B locally?
Yes — Step 3.5 Flash REAP 121B A11B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Step 3.5 Flash REAP 121B A11B?
At Q4_K_M, Step 3.5 Flash REAP 121B A11B can reach ~94 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)
Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.
Example: NVIDIA B200 → 7.9 GB active ÷ (8000 × 0.65) = 1.52 ms, plus 45 layers × 0.055 ms = 2.48 ms, so 1000 ÷ 4.00 ms = ~250 tok/s
Estimated speed at Q4_K_M (79.8 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Step 3.5 Flash REAP 121B A11B?
At Q4_K_M, the download is about 72.58 GB. The full-precision BF16 version is 241.93 GB. The smallest option (Q2_K) is 51.41 GB.
Which GPUs can run Step 3.5 Flash REAP 121B A11B?
No single consumer GPU has enough VRAM to run Step 3.5 Flash REAP 121B A11B at Q4_K_M (79.8 GB). Multi-GPU or professional hardware is required.
Which devices can run Step 3.5 Flash REAP 121B A11B?
30 devices with unified memory can run Step 3.5 Flash REAP 121B A11B at Q4_K_M (79.8 GB), including ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB), GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.