GPT-OSS 120B
GPT-OSS 120B is an open-source large language model released by OpenAI with 120 billion parameters, making its weights available to developers and researchers. The model provides strong natural language understanding and generation capabilities, supports fine-tuning and deployment for various downstream tasks, and balances high performance with openness, allowing free use and integration across multiple platforms and environments.
Core Facts
Timeline (last 90 days)
Developers tried using GPT-OSS 120B to propose a retraining curriculum; in a matched development pilot on one seed, this approach performed worse than a simple fixed 5% stress policy, and neither passed the nominal performance preservation guardrail.
GPT-OSS 120B failed on the e5 information extraction task (the height of the Eiffel Tower), while all other models passed.
The models tested include Google's Gemini 3.7 Flash, Z.ai's GLM-5, Alibaba's Qwen 3 235B A22B Instruct, and OpenAI's gpt-oss-120b, all running on Kaggle.
The author selected the free model GPT-OSS 120B hosted on the Groq platform as the chat model in n8n's AI Agent.
OpenAI used Semi-Analysis's public benchmark InferenceX to test Jalapeño on three open models: GPT-OSS 120B, DeepSeek R1 670B, and Kimi K2 1T.
In the gpt-oss 120B test, Jalapeño achieved a peak throughput of approximately 85,448 per watt compared to GB200's 44,960, roughly 1.9×; end-to-end latency was reduced from 1.80 seconds to 1.03 seconds; and minimum TBT dropped from 1.87 ms to 0.69 ms.
The Jalapeño benchmark simultaneously ran three models employing different MoE and attention mechanisms: GPT-OSS 120B, DeepSeek R1, and Kimi K2.5.
Jalapeño peaks at 1500 tokens/second when running the GPT-OSS 120B model, approximately twice that of the next-best solution.
On the 120-billion-parameter GPT-OSS 120B, token generation improves from 90 to 130 tokens/s, and PP improves from approximately 1,300 to approximately 3,200 tokens/s.
All Facts (10)
GPT-OSS 120B需要大于80GB显存（如H100 GPU）才能运行
85%Unverified单卡RTX 4090即可运行GPT-OSS 20B，单卡A100即可运行GPT-OSS 120B
80%Partially VerifiedThe models tested include Google's Gemini 3.7 Flash, Z.ai's GLM-5, Alibaba's Qwen 3 235B A22B Instruct, and OpenAI's gpt-oss-120b, all running on Kaggle.
65%UnverifiedIn the gpt-oss 120B test, Jalapeño achieved a peak throughput of approximately 85,448 per watt compared to GB200's 44,960, roughly 1.9×; end-to-end latency was reduced from 1.80 seconds to 1.03 seconds; and minimum TBT dropped from 1.87 ms to 0.69 ms.
60%UnverifiedDevelopers tried using GPT-OSS 120B to propose a retraining curriculum; in a matched development pilot on one seed, this approach performed worse than a simple fixed 5% stress policy, and neither passed the nominal performance preservation guardrail.
50%UnverifiedOpenAI used Semi-Analysis's public benchmark InferenceX to test Jalapeño on three open models: GPT-OSS 120B, DeepSeek R1 670B, and Kimi K2 1T.
50%UnverifiedThe Jalapeño benchmark simultaneously ran three models employing different MoE and attention mechanisms: GPT-OSS 120B, DeepSeek R1, and Kimi K2.5.
50%UnverifiedJalapeño peaks at 1500 tokens/second when running the GPT-OSS 120B model, approximately twice that of the next-best solution.
50%UnverifiedOn the 120-billion-parameter GPT-OSS 120B, token generation improves from 90 to 130 tokens/s, and PP improves from approximately 1,300 to approximately 3,200 tokens/s.
50%Unverified使用普通 Google 账号登录 Antigravity 可免费调用 Gemini 3 Pro、Gemini 3.5 Flash、Claude Opus 4.6 Thinking 和 Sonnet 4.6 等模型
50%