gpt-oss-120b

Introduction

gpt-oss-120b is OpenAI’s first open-weight language model, released in August 2025 under a permissive Apache 2.0-style license. It is a 117-billion-parameter mixture-of-experts model with about 5.1 billion active parameters per forward pass, built for high-reasoning, agentic, and general-purpose production use, and efficient enough to run on a single high-end GPU. Like other foundation models on this site, it is reviewed here as raw AI infrastructure that legal teams and vendors can build on via API or self-hosting, not as a packaged legal product.

Legal Research

gpt-oss-120b has strong general reasoning and tool-calling ability, which can support research workflows such as summarizing case law fed into its context or orchestrating searches against an external legal database. It has no native connection to Westlaw, Lexis, or any legal citation source, and no built-in citation verification, so any research pipeline built on it needs an added retrieval and verification layer to be trustworthy for legal use.

Contract Analysis

With appropriate prompting, gpt-oss-120b can extract clauses, flag risk language, and compare contract text to a template, and its strong tool-calling and coding benchmarks suggest it integrates well into automated review pipelines. There is no purpose-built contract-review interface bundled with the model, so firms need custom tooling or a third-party product built on top of it.

Document Drafting

The model performs competitively on general problem-solving, coding, and math benchmarks (matching or beating OpenAI's own o4-mini on several), which translates into capable drafting of contracts, memos, and correspondence from prompts or templates. It ships without a legal-specific drafting interface, template library, or word-processor integration.

Legal Translation

As a general-purpose multilingual model, gpt-oss-120b can translate and summarize legal text across languages, but it is not tuned specifically for legal terminology or jurisdiction-specific concepts, so translations should be checked by a qualified bilingual reviewer for anything client-facing.

Pricing

gpt-oss-120b is free to download and self-host under its open-weight license, and is efficient enough to run on a single H100-class GPU. Through managed API providers, 2026 pricing is very low compared to closed frontier models — ranging from roughly $0.04 to a few dollars per million tokens depending on the host and quantization — though, as with any open-weight model, legal teams must add their own cost for hosting, fine-tuning, and any legal-specific tooling built around it.

User Reviews and Ratings

As a developer-facing foundation model, gpt-oss-120b does not have law-firm-oriented reviews on sites like G2. Independent 2026 benchmarks show it outperforming OpenAI's own o3-mini and matching or exceeding o4-mini on competition coding, tool use, and math, with developer commentary highlighting its strong price-to-performance ratio and single-GPU efficiency as its main advantages.

Verdict

gpt-oss-120b is a strong, low-cost open-weight option for teams with the technical capacity to build legal-specific tooling on top of it, offering performance close to OpenAI's proprietary small models at a fraction of the cost. It is not a ready-to-use legal assistant, so firms wanting an out-of-the-box research, drafting, or contract-review product should look at purpose-built legal AI tools instead, whether or not those tools happen to use gpt-oss under the hood.