GPT-5 Nano

Portkey describes GPT-5 Nano as the smallest and most efficient model in OpenAI's GPT-5 lineup, engineered for ultra-low latency and high throughput production workloads. It is positioned for real-time use cases including chatbots, coding assistants, and multi-agent systems where response time and cost matter more than frontier reasoning. The model inherits GPT-5 family system improvements including faster inference paths, improved grounding, and more stable outputs across conversations. Portkey notes Nano trades deep reasoning and contextual nuance for efficiency, and developers can reserve GPT-5 or GPT-5 Mini for complex reasoning while slotting Nano into latency-sensitive production paths.

GPT-5 Nano from OpenAI was released on August 7, 2025 as a compact model in the GPT-5 lineup. According to Vals AI, it supports a 400k token context window with 128k max output tokens, text and image input, and pricing at $0.05 per million input tokens and $0.40 per million output tokens. Default provider listed is OpenAI. Benchmark scores reported by Vals AI include 63.38% on GPQA Diamond, 70.22% on LiveCodeBench, 76.07% on MMLU Pro, and 50.13% on LegalBench, with an overall accuracy of 58.54%. Average latency measured at 55.01 seconds. The model ranks competitively across academic and industry benchmarks against other models in the Vals index.

Simon Willison's coverage of GPT-5 from preview access confirms the API exposes three variants—regular, mini, and nano—each runnable at minimal, low, medium, or high reasoning levels. Nano is described alongside mini as part of the unified, tiered API offering rather than a separate product. GPT-5 in the API supports 272K input tokens and 128K output tokens, with text and image input but text-only output, applying across all variants including nano. Once usage limits are reached in ChatGPT, a mini version handles remaining queries, framing nano's role in the fallback chain.

GPT-5 Nano is OpenAI's smallest, fastest variant in the GPT-5 system, according to OpenRouter's listing, succeeding GPT-4.1-nano for cost-sensitive and real-time use. It was released on August 7, 2025 with a 400K context window and a knowledge cutoff of May 2024. List pricing is $0.05 per 1M input tokens and $0.40 per 1M output tokens, with Azure and OpenAI both reaching 100% uptime on the standard route. OpenRouter Flex pricing averages around $0.025 input and $0.20 output per 1M tokens with strong token-share throughput.