OpenAI
OpenAI
ReasoningOpen Weights (Open) Verified Architecture & SpecsFree ($0 API Tokens)

GPT OSS 120B

GPT OSS 120B is a reasoning model from OpenAI with 117B parameters, supporting a 131,072-token context window, with text modalities. Open-weight for self-hosted or compatible deployments.

Technical Architecture & Execution Specifications
Architecture Overview

GPT OSS 120B

GPT OSS 120B is a reasoning model from OpenAI with 117B parameters, supporting a 131,072-token context window, with text modalities. Open-weight for self-hosted or compatible deployments.

Memory Math Breakdown
  • • FP16 Weights = 117.0B × 2B = 234.00 GB
  • • INT4 Weights = 117.0B × 0.55B = 64.35 GB
  • • KV Cache (131072 ctx, FP16) ≈ 80.00 GB
  • • Activation Buffer = ~20% overhead
Supported Modalities
text

Hardware & Execution ParametersReasoning

Total Parameter Count117B
Active Parameters (MoE)5.1B per token
Context Window Capacity131,072 tokens
Model Weights Footprint234.0 GB (FP16) / 64.4 GB (INT4)
Distribution LicenseOpen Weights (Open)
Standard API Pricing (1M Tokens)Free / Self-Hosted
Model Heritage & Evolutionary Lineage

Genealogical Graph & Evolutionary Provenance

Tracing foundational base architecture ancestry, architectural successors, scale siblings, and reasoning distillation derivatives.

Independent Foundation Checkpoint

GPT OSS 120B operates as an autonomous foundation model architecture without direct precursor derivatives in this catalog.

Root Architecture Node
AI Model Architecture & Intelligence

Architecture Engineering & Capability Deep-Dive

An objective architectural evaluation of GPT OSS 120B by OpenAI, analyzing underlying compute dynamics, memory constraints, and deployment economics.

Topology & Attention Mechanics

Features an Omni multimodal unified encoder capable of test-time compute scaling via explicit Chain-of-Thought (CoT) reasoning tokens.

Architecture Type:Advanced Transformer

Evaluation Profile & Reasoning

Exhibits frontier-tier behavior in reasoning and coding. World-class step-by-step mathematical extraction.

Domain Specialty:Reasoning

LLM Hardware Sizing & Serving

For open deployments via vLLM/SGLang, quantization (INT4/AWQ) is heavily recommended to fit dense memory constraints, or multi-GPU pipeline parallelism for full FP16.

KV Cache Mgmt: PagedAttention / FlashAttention-3
Hosting Type:Open Weights (Open)

Inference Economics & Workflows

Reasoning tokens dynamically scale compute on hard problems. Expect higher output costs and varied TTFB, offset by massive reductions in hallucination rates.

Enterprise Fit:Production Ready
Production Trade-Offs & Capability Balance

Architectural Strengths vs. Considerations

An objective balance sheet analyzing the operational advantages and production constraints of deploying GPT OSS 120B.

Key Architectural Strengths

  • Omni multimodal unified encoder with test-time compute scaling (CoT reasoning tokens) maxes out complex problem solving.
  • Exceptional adherence to structured JSON schemas accelerates integration into deterministic enterprise pipelines.
  • Massive 131,072-token context allows full-repository and book-length ingestion.
  • Demonstrated AIME 2024 evaluation score of 95.8% in verified benchmarks.

Operational Considerations

  • Autoregressive CoT reasoning tokens can increase Time-to-First-Byte (TTFB) and inflate output token budgets unpredictably.
  • 128k+ token prefill stages become heavily compute-bound and balloon KV cache without PagedAttention chunking.
LLM Hardware Sizing & Runtime Compatibility

Inference Runtimes & Hardware Sizing

Deployment targets, inference engines, and memory requirements for GPT OSS 120B.

Recommended Hardware Profile:
4x-8x H100 (80GB) with Tensor Parallelism
Multi-Node / Multi-GPU Cluster (>160 GB)
Est. 280.8 GB (FP16) / 77.2 GB (INT4)
Production Serving Recipes
vLLM Production
python3 -m vllm.entrypoints.openai.api_server --model GPT OSS 120B --tensor-parallel-size 4 --gpu-memory-utilization 0.95 --max-model-len 4096 --enable-chunked-prefill
Ollama / llama.cpp
ollama run gpt oss 120b
SGLang Structured
python3 -m sglang.launch_server --model-path GPT OSS 120B --tp 4 --trust-remote-code
TGI Serving
text-generation-launcher --model-id GPT OSS 120B --num-shard 4 --max-batch-prefill-tokens 32000
Supported Inference Engines
vLLM

High-throughput PagedAttention server

Supported
Ollama

One-click CLI & local desktop serving

Supported
SGLang

Fast multi-turn structured decoding

Supported
TGI

Text Generation Inference

Supported
Llama.cpp

GGUF CPU/Apple Silicon execution

Supported
Precision & Quantization Formats
BF16 / FP16Full Precision

~280.8 GB VRAM required

FP8 (E4M3)Native FP8

~140.4 GB VRAM (Hopper speedup)

AWQ / GPTQ (4-bit)Activation-Aware

~77.2 GB VRAM

GGUF (Q4_K_M)Quantized Binary

CPU RAM / Apple Silicon optimized

LLM Benchmark Database & Performance Metrics

12 Tested

Standardized evaluation results across reasoning, agentic coding, computer use, and alignment.

Flagship Headline MetricsIndustry SOTA Standard
Reasoning & Science

GPQA Diamond

80.1%accuracy
0%100%
Reasoning & Science

GPQA Diamond

80.9%accuracy
0%100%
Coding & Software

SWE-Bench Verified

62.4%resolve rate
0%100%
AIME 2024
accuracy
95.8%
AIME 2024
accuracy
96.6%
AIME 2025
accuracy
92.5%
AIME 2025
accuracy
97.9%
GPQA Diamond
accuracy
80.1%
GPQA Diamond
accuracy
80.9%
Humanity's Last Exam
accuracy
14.9%
Humanity's Last Exam
accuracy
19%
Commercial Rates & Inference Costs

API & Deployment Pricing

Open-weights model available for local and private cloud deployment. Compute costs depend on the target GPU hardware instance.

Deployment TierPricing Structure
Open Checkpoint Weights$0.00 (Free Download)
Inference Token Consumption$0.00 / Token
Inference Cost & ROI Engine
Market Cloud Rate
Prompt / Input Volume:50M Tokens / mo
1M500M1,000M
Generated / Output Volume:10M Tokens / mo
1M250M500M
Estimated Monthly Spend
$160.00/ mo
Input (50M @ $2.00/1M):$100.00
Output (10M @ $6.00/1M):$60.00
Cloud GPU Breakeven Ratio
RunPod RTX 4090 ($316/mo)0.51x spend
Lambda 1x H100 ($1,800/mo)0.09x spend
💡 Open-weights model. You can self-host for $0 token API charge or consume via managed serverless endpoints at the rates shown above.
Similar Frontier Models & Alternatives
Explore All Comparisons

Comparable Foundation Architectures

Alternative models in the Reasoning class with similar capabilities, context windows, or deployment profiles.

TencentReasoning

Hy4 preview

Context:1024k ctx
Parameters:Open Weights
Input Rate:Free / Self-Host
InclusionaiReasoning

Ling 3.0 Flash Fin

Context:262k ctx
Parameters:Proprietary
Input Rate:Free / Self-Host
Alibaba CloudReasoning

Qwen3.8-Flash-Next

Context:262k ctx
Parameters:176B
Input Rate:$0.15/1M
Integration & Deployment
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("OPENAI_KEY", "EMPTY"),
    base_url="http://localhost:8000/v1"
)

response = client.chat.completions.create(
    model="openai-gpt-oss-120b",
    messages=[{"role": "user", "content": "Explain quantum superposition in 2 sentences."}]
)
print(response.choices[0].message.content)
Frequently Asked Questions

Frequently Asked Questions about GPT OSS 120B

Essential facts, architectural specs, hardware constraints, and pricing answers for GPT OSS 120B.

To run GPT OSS 120B (117B) locally, you generally need 8-12 GB VRAM (4-bit to 8-bit). We recommend using quantized GGUF/AWQ formats with Ollama or vLLM to optimize memory footprint.

Primary Sources & Access Repositories

All technical specifications, parameter distributions, context architectures, and benchmark evaluations for GPT OSS 120B are audited against primary source release documentation, research whitepapers, and verified vendor API endpoints.

api id:openai/gpt-oss-120b