gpt-oss-120b

gpt-oss-120b is an open-weight language model released by OpenAI as part of its gpt-oss family, built on a Mixture-of-Experts architecture with 117 billion total parameters and 5.1 billion activated per forward pass. Native MXFP4 quantization lets the weights run efficiently on a single H100 GPU, lowering the barrier to self-hosted or high-throughput deployments. The model is trained in OpenAI's Harmony response format and exposes configurable reasoning depth along with full chain-of-thought access, giving developers explicit control over how much deliberation happens before each answer.

Beyond text generation, gpt-oss-120b is designed for agentic and production workloads, with native support for function calling, browsing, and structured output generation that fits into pipelines requiring JSON-constrained responses or external tool orchestration. Third-party listings report strong results on graduate-level reasoning evaluations such as GPQA Diamond, alongside competitive scores on instruction-following and conversational agent benchmarks, indicating a balanced profile rather than a narrow specialist. Open weights, a long context window, and adjustable reasoning effort make it a practical fit for teams that want a transparent reasoning model they can fine-tune or self-host while still benefiting from OpenAI-aligned agentic capabilities.

Privatemode AIgpt-oss-120bgpt-oss

Open provider page Provider docs

Quick Info

Provider: Privatemode AI
Model key: gpt-oss-120b
Release date: Aug 5, 2025
Last updated: Aug 5, 2025
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.2311
Output token cost: $0.7511

Limits

Output tokens: 32,768 tokens
Context window: 128,000 tokens

Transparent token rates

Compare gpt-oss-120b pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about gpt-oss-120b

Privatemode AI

CoverageRelease Notes

Discover more about what's new at AWS with OpenAI GPT OSS and NVIDIA Nemotron Models Available on Amazon Bedrock in AWS GovCloud (US)

Privatemode AI

CoverageComparison

8x NVIDIA GB10 GPT OSS 120B Throughput Vs TP Width. 8x NVIDIA GB10 Cluster Monitoring Under Load. 8x NVIDIA GB10 GPT OSS 120B Throughput Vs TP Width.

Privatemode AI

CoverageComparison

8x NVIDIA GB10 GPT OSS 120B Concurrency Vs TP. 8x NVIDIA GB10 Cluster Monitoring Under Load. 8x NVIDIA GB10 GPT OSS 120B Concurrency Vs TP.

Privatemode AI

Coverage

New prompt-based policies aim to standardise how AI systems protect under-18 users.

Privatemode AI

Coverage

As AI agent deployments grow across enterprise systems, organizations need a way to bring together foundation models, tools, and governance frameworks to...

Privatemode AI

Coverage

HyperNova 60B 2602, a 50% compressed version of OpenAI’s gpt-oss-120B, accelerates Multiverse’s plans to deliver hyper-efficient, high-performance models for free to developersDONOSTIA, Spain, Feb. 24, 2026 (GLOBE NEWSWIRE) -- Multiverse Computing, the leader in AI model compression, today announced the release of Hype

Regolo AI

CoverageBenchmark

An arXiv preprint (December 2025) by Bi, Zhang, Song, and Tseng benchmarks the GPT-OSS model family against contemporary LLMs across ten financial NLP tasks, including sentiment analysis, question answering, and entity recognition on datasets such as Financial PhraseBank, FiQA-SA, and FLARE FINER-ORD. The evaluation co The study reports a counterintuitive finding: GPT-OSS-20B achieves accuracy of 65.1% versus 66.5% for GPT-OSS-120B, while delivering a 198.4 Token Efficiency Score and processing at 159.80 tokens per second. The authors conclude that GPT-OSS models consistently outperform larger competitors including Qwen3-235B, sugges

Weights & Biases

CoverageRelease Notes

NVIDIA NIM for Large Language Models Release 2.0.11 officially documents several deployment constraints specific to OpenAI's gpt-oss-120b. For the MXFP4 TP8 profile on NVIDIA A10G and NVIDIA A100-SXM4-40GB, the release notes require NIM_KVCACHE_PERCENT=0.80 so that the sampler warm-up has sufficient device memory. Addi The fused mixture-of-experts kernels shared by gpt-oss-120b, gpt-oss-20b, and nemotron-3-nano support LoRA adapters with a maximum rank of 128, with adapters above that rank explicitly unsupported. The NIM 2.0.11 release also updates the inference backend to vLLM 0.27.0 and ships SGLang Model-Free NIM container v2.1.2.