OpenAI
OpenAI: gpt-oss-120b
openai/gpt-oss-120b
gpt-oss-120b is an open-weight, 117B-parameter Mixture-of-Experts (MoE) language model from OpenAI designed for high-reasoning, agentic, and general-purpose production use cases.
- Context window: 131,072 tokens
- Input: text
- Output: text
- Pricing: $0.15/M input tokens, $0.6/M output tokens
Community sentiment
Mixed (+10) — 6 positive, 5 negative, 6 neutral across 17 community samples from X and the web, as of 2026-10-27.
Community commentary portrays gpt-oss-120b as a fast, low-cost open-weight model valued for single-GPU efficiency, agent tasks, extraction and high request caps, but users note it is weaker on hardest reasoning benchmarks and has seen removal from some free/paid tiers. Benchmarks and reports show mixed results — top throughput and price in some tests, but failures or lower accuracy in others.
- “334 tok/s and $0.10/M: gpt-oss-120b is the fastest open-weight model on https://t.co/5P7vJMSMwe today, and nearly the cheapest. Open-weight top 5 by output speed (best provider, 24h median), with the cheapest blended price: 1. gpt-oss-120b: 334 tok/s, $0.10/M 2. GLM 5.3: 278 tok/” — X
- “Groq: 30 RPM, 1K requests/day, 8K TPM, 200K tokens/day on openai/gpt-oss-120b. Very fast inference — good for latency-sensitive side projects.” — X
- “926 Steps, Two Lobotomies, and Zero Nudges: Flight-Testing the ... — 5 days ago · AI agents write 30% more code and close zero additional tickets ・ a 120B open-weight model fell right through it: GPT-OSS 120B ・ the LLM is just a swappable ...” — the web
- “@0xzeiv gpt-oss-120b on both. open weights, 120B great for agents, extraction and bulk edits. not for your hardest reasoning task free is the volume lane, not the frontier lane” — X
RuntimeWire coverage
- Head to head: DeepSeek-V4-Pro vs gpt-oss-120b
- Head to head: Kimi-K3 vs gpt-oss-120b
- Cerebras details CS-4 rack and targets 10,000 tokens per second with CS-5
- Ox Alpha filters domestic Chinese political risks, CTGT audit finds
- Inco AI releases DFlash 2, reports 21% longer accepted token sequences
- Head to head: DeepSeek-V4-Flash-0731 vs gpt-oss-120b
- Head to head: gpt-oss-120b vs Phi-4-reasoning
- Head to head: gpt-oss-120b vs cohere-command-a
- CTGT says DeepSeek distillation lifted finance scores without censorship transfer
- Head to head: Google: Gemini 3.6 Flash vs gpt-oss-120b
- Head to head: gpt-oss-120b vs Kimi K3
- Head to head: DeepSeek-V4-Pro vs gpt-oss-120b
View on OpenRouter. Model data sourced from OpenRouter.