Open weightsOpenAI/10 trustVerified

GPT-OSS 120B

Name: GPT-OSS 120B
Rating: 9

OpenAI's first open-weights model — 117B MoE reasoning under Apache 2.0

Visit official page →

About

gpt-oss-120b is OpenAI's first open-weights model, released August 5, 2025 alongside gpt-oss-20b. It is a mixture-of-experts reasoning model with 117B total parameters and 5.1B active per token (128 experts, top-4), a 128K-token context window, and native MXFP4 quantization that lets it run on a single 80GB GPU. Weights are on Hugging Face under Apache 2.0 with no usage restrictions; OpenAI does not host an API for it, so inference comes from third-party providers or self-hosting.

Specs

Context window

128K tokens

Pricing

$0.15 input / $0.59 output per 1M tokens (Artificial Analysis median across providers; e.g. $0.15/$0.75 via Groq)

Free quota

Open weights free to download and self-host; no OpenAI-hosted API; some third-party hosts offer free tiers (unverified)

API access

No OpenAI-hosted API; third-party hosts only (Groq, Together, Fireworks, SambaNova, Novita, OpenRouter, etc.); local via Ollama, vLLM, llama.cpp

Weights

Open weights

Released

August 5, 2025

Benchmarks

Near-parity with o4-mini on core reasoning; outperforms o3-mini — provider-reported
Matches/exceeds o4-mini on Codeforces, MMLU, HLE, TauBench; exceeds o4-mini on HealthBench and AIME 2024/2025 — provider-reported
Artificial Analysis Intelligence Index: 12 (high) / 10 (low), above the open-weight median of 8 — independent
239 tokens/sec output, 0.84s time-to-first-token (AA median across providers) — independent
AIME 2025 93.4%, LiveCodeBench 87.8% — third-party aggregator (Requesty; methodology unclear)

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

reasoningchatcodedatasearch

Trust score — 9/10

/10 trustVerified

Existence, Apache 2.0 license, release date (2025-08-05), architecture and AA independent scores all cross-verified (OpenAI announcement, HF blog, Artificial Analysis). Minor caveats: provider "near-parity with o4-mini" claim looks generous next to the independent AA index of 12; text-only, no vision; no OpenAI-hosted API. Metrics-v1 score 9/10 — adoption 10, maintenance 5, openness 10, manual 9.

How the trust score is built

Full methodology →

Adoption · stars/downloads, log-scaled vs peers: 10.0/10
Maintenance · recency, releases, issue load: 5.0/10
Openness · license + docs: 10.0/10
Manual review · sanity read + red-flag check: 9.0/10

· HF top match: openai/gpt-oss-120b
· HF downloads 3,644,353 (peer-normalized)
· released 434 days ago
· manual review: 'near-parity with o4-mini' looks generous vs independent AA score; text-only