gpt-oss-120b · 3 GPUs measured first-party · llama.cpp · Updated October 2026
What GPU Do You Need for gpt-oss-120b?
gpt-oss-120b on 3 GPUs, measured first-party: NVIDIA RTX PRO 6000 Blackwell Workstation Edition leads at 253.9 tok/s, A100 80GB SXM4 trails at 143 tok/s, and it peaked at 64GB of VRAM.
Benchmarked weights: ggml-org/gpt-oss-120b-GGUF
NVIDIA RTX PRO 6000 Blackwell Workstation Edition
253.9 tok/s on gpt-oss-120b, the ceiling. Measured on our bench. 96GB of VRAM, $8,565 at launch.
- 253.9 tok/s on gpt-oss-120b
- 96GB, clears the gpt-oss-120b floor
- Rentable by the hour rather than bought
- 600W board rating
- Datacenter or workstation hardware, not a retail purchase
What GPU Do You Need for gpt-oss-120b?, tok/s by GPU
Efficiency: tok/s per 100W drawn
Power is the average pulled during the run, sampled at 1Hz. The fastest card is often not the one here, and for anything left running this is the number that shows up on the bill.
Value: tok/s per $1,000 of MSRP
Launch price, not street price, so it ages. A speed leaderboard always crowns the most expensive card; this is the counterweight.
gpt-oss-120b. Measured tokens per second by GPU
| GPU | tok/s | Prompt t/s | tok/W | Avg power |
|---|---|---|---|---|
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 253.9 | 7876 | 1.81 | 140.0 W |
| NVIDIA H100 80GB HBM3 | 220.4 | 3984.9 | 1.43 | 154.2 W |
| NVIDIA A100 80GB SXM4 | 143.1 | 1756.7 | 1.25 | 114.7 W |
What the numbers show. Across 3 GPUs measured on our own bench, RTX PRO 6000 Blackwell Workstation Edition is fastest at 254 tok/s. The slowest, A100 80GB SXM4, manages 143, so the spread is 1.8x from top to bottom.
About gpt-oss-120b. gpt-oss-120b: from openai, 117B parameters, on Hugging Face since August 2025, Apache 2.0 licence. 4,188,657 downloads in the last 30 days.
How it compares. H100 80GB HBM3: gpt-oss-120b 220.4 tok/s, Qwen3-Coder-Next 193.8 (80B), Qwen2.5-72B 40.78 (73B), Qwen3.6 35B A3B 236.7 (36B), Ornith 1.5 35B A3B 240.2 (36B). 2 of 4 beat gpt-oss-120b here. Qwen3.6 35B A3B and Ornith 1.5 35B A3B are mixture-of-experts, so per token they compute only a slice of their size.
Cost on a rented GPU. 1M generated tokens of gpt-oss-120b: $1.18 on a RTX PRO 6000 Blackwell Workstation Edition ($1.08/hr, 66 min).
gpt-oss-120b: cost per 1M generated tokens on rented GPUs
| GPU | Cheapest rate | Speed (tok/s) | Cost per 1M generated tokens |
|---|---|---|---|
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | $1.08/hr | 253.9 | $1.18 |
| NVIDIA A100 80GB SXM4 | $0.95/hr | 143.1 | $1.84 |
| NVIDIA H100 80GB HBM3 | $2.14/hr | 220.4 | $2.69 |
Cheapest hourly rate we track on RunPod and Vast.ai, divided by the measured speed. Startup time and storage are extra.
Speed tiers for gpt-oss-120b. 30+ tok/s: 3 (RTX PRO 6000 Blackwell Workstation Edition, H100 80GB HBM3, A100 80GB SXM4). 30 tok/s is roughly where replies outpace reading.
Reading your prompt. Before gpt-oss-120b writes anything it reads the input: 7876.0 tok/s on the RTX PRO 6000 Blackwell Workstation Edition (0.5s for a 4,000-token prompt), 1756.7 on the A100 80GB SXM4 (2.3s). Long documents and big code files feel this number more than the generation speed.
VRAM for gpt-oss-120b. Measured peak 60.0GB, so 80GB is the smallest common card size; smallest card it ran on: H100 80GB HBM3 (80GB).
Power on gpt-oss-120b. Most efficient: RTX PRO 6000 Blackwell Workstation Edition, 140W, 0.15 kWh per 1M generated tokens. Hungriest: H100 80GB HBM3, 154W, 0.19 kWh. At $0.15/kWh: $0.023 per 1M generated tokens.
Our verdict
Fastest on gpt-oss-120b: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, 253.9 tok/s. Cheapest to rent per job: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, $1.18 per 1M generated tokens.
FAQ
- What GPU do I need to run gpt-oss-120b?
- About 63GB. Smallest card that ran it: NVIDIA H100 80GB HBM3 (80GB).
- How much does it cost to run gpt-oss-120b in the cloud?
- $1.18 per 1M generated tokens on a NVIDIA RTX PRO 6000 Blackwell Workstation Edition at $1.08/hr, cheapest of 3 rentable cards we measured.
- Can I run gpt-oss-120b on a 12GB, 16GB or 24GB card?
- It used 60.0GB at the precision we tested. 12GB: no; 16GB: no; 24GB: no.
- Is the H100 80GB HBM3 or the A100 80GB SXM4 faster for gpt-oss-120b?
- The H100 80GB HBM3: 220.4 vs 143.1 tok/s, 54% faster on our bench.
How we test
llama.cpp llama-bench at Q4_K_M, 512-token prompt and 128 generated tokens, three runs after a warmup, full GPU offload, with power and VRAM sampled throughout. Token generation is memory-bandwidth-bound, so the ranking tracks bandwidth closely, which makes it a fair guide to cards we haven't run yet.