Artificial Intelligence

OpenAI shrinks GPT-5.4 for speed

OpenAI unveils GPT-5.4 mini and nano, smaller, faster AI models optimized for coding and subagent tasks, offering improved efficiency and lower costs.

StartupHub.ai Staff

By StartupHub.ai Staff, reviewed by Daniel Singer, Editor

3 min read

OpenAI shrinks GPT-5.4 for speed
Credit: OpenAI News
Contents(4)

OpenAI is expanding its GPT-5.4 family with the introduction of two new, more compact models: GPT-5.4 mini and GPT-5.4 nano. These releases, detailed on OpenAI News, are designed for efficiency and speed, targeting applications where rapid response times and cost-effectiveness are paramount.

GPT-5.4 mini represents a substantial upgrade over the previous GPT-5 mini, delivering enhanced performance in coding, reasoning, and multimodal understanding. Notably, it runs over twice as fast and approaches the benchmarks of the larger GPT-5.4 on tasks like SWE-Bench Pro and OSWorld-Verified.

The GPT-5.4 nano model is positioned as the most economical and swift option within the GPT-5.4 series. OpenAI recommends it for tasks such as classification, data extraction, and powering simpler subagents, where minimal latency and cost are key drivers.

Optimized for Action

These new models are engineered for workloads where responsiveness directly impacts user experience. This includes coding assistants demanding near-instantaneous feedback, subagents executing routine tasks quickly, and computer-using systems that process screenshots in real-time.

The company emphasizes that for many applications, the most effective model isn't necessarily the largest. Instead, it's the one that balances speed, reliable tool usage, and strong performance on complex professional tasks.

Performance Benchmarks

Early benchmarks indicate GPT-5.4 mini's strengths across various domains. In coding tasks, it significantly outperforms GPT-5 mini at similar latencies, nearing GPT-5.4 levels of success while operating much faster. This offers a compelling performance-per-latency trade-off for developers.

GPT-5.4 mini also shows promise in multimodal applications, particularly those involving computer vision. It can swiftly interpret complex user interfaces from screenshots, demonstrating performance close to GPT-5.4 on the OSWorld-Verified benchmark.

Subagents and Compositional AI

The introduction of these smaller, faster models also facilitates more sophisticated AI architectures. OpenAI highlights their use in subagent systems, where larger models can orchestrate tasks, delegating specialized sub-functions to GPT-5.4 mini or nano for rapid execution.

This compositional approach, where developers combine models of varying sizes for optimal efficiency, becomes increasingly viable as smaller models gain capability. It allows for systems where a central model handles planning, while smaller, specialized models execute specific tasks at scale.

Availability and Pricing

GPT-5.4 mini is now available via the OpenAI API, Codex, and ChatGPT. The API version supports text and image inputs, tool use, and boasts a 400k context window, priced at $0.75 per 1M input tokens and $4.50 per 1M output tokens.

GPT-5.4 nano is exclusive to the API, offered at a more aggressive price point of $0.20 per 1M input tokens and $1.25 per 1M output tokens, making it suitable for high-volume, cost-sensitive applications.

Mentioned in this story

Sources

Get the free daily brief every morning

Every morning the day's AI news, new AI startups and new funding rounds. Free and takes a few minutes to read, unsubscribe at any time.

Join 64,000 other subscribers.

TopicsOpenAIGPT-5.4AI ModelsMachine LearningLLMCoding AISubagents

StartupHub.ai StaffDaniel SingerWritten by StartupHub.ai Staff, Editorial teamReviewed by Daniel Singer, EditorAbout

StartupHub.ai Staff, Editorial team

The staff writers of StartupHub.ai, ranging from investment analysts to avid AI tool users, early adopters and critical enthusiasts. Backgrounds span engineering, business and the arts. We hold every piece to rigorous standards of research and review.

Daniel Singer, Editor

Daniel Singer is an AI expert who has been writing about, covering and building AI technology since 2018.

More from Daniel Singer

Test your Agent Readiness now

Paste your website. How well ChatGPT, Claude and the other agents can read and use a site, in about a minute.

Watch

The videos behind the news, one flip at a time

The feed mixes the videos from our articles with startup demos and you flip through them like TV channels for free with no signup.