GPT-5.4 Mini
GPT-5.4 Mini by OpenAI is a fast, efficient small model optimized for coding, subagent delegation, computer use, and multimodal reasoning at 400k context.
Model Overview
Capabilities, design details, and architectural traits
GPT-5.4 Mini - Distilled Speed for Subagents and Coding Workflows
GPT-5.4 Mini is OpenAI's distilled small model built to bring GPT-5.4's core strengths - coding, tool use, multimodal reasoning, and computer use - to a faster form. It runs more than 2x faster than GPT 5 Mini.
Designed for the Subagent Layer
GPT-5.4 Mini is explicitly positioned as the execution layer in multi-model pipelines. In Codex, for example, a larger model like GPT-5.4 handles planning and coordination, while GPT-5.4 Mini runs as a subagent completing narrower parallel tasks - codebase search, file review, document processing. This division is built into how Codex allocates quota: GPT-5.4 Mini uses only 30% of the GPT-5.4 quota per task.
What Separates It
| Trait | Detail |
|---|---|
| Subagent-native design | Officially described as the target model for parallel subagent execution inside Codex multi-model workflows |
| 400k context window | Handles large codebase navigation and long document tasks at the mini tier |
| Multimodal reasoning | Supports text and image inputs with real-time image interpretation as a documented primary use case |
| Tool use reliability | Supports function calling, web search, file search, computer use, and skills natively via the /v1/responses endpoint |
Available in Three Surfaces
GPT-5.4 Mini is deployed across the API, Codex (app, CLI, IDE extension, web), and ChatGPT (as the Thinking option for Free and Go users, and as a rate-limit fallback for other tiers). This cross-surface availability is unique among mini-tier models in the GPT-5.4 family - GPT-5.4 Nano, by contrast, is API-only.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | OpenAI | Anthropic | Anthropic |
| Release Date | March 17, 2026 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | Aug 2025 | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 400K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | $0.75 Best Input Pricing | $4 | $2 |
| Output Pricing | $4.50 Best Output Pricing | $20 | $10 |
| Modalities | |||
| Inputs | fileimagetext | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 19.7 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | - | - | - |
| Agentic Index | - | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from OpenAI
Other models by OpenAI
Top AI Models
Leading alternatives by intelligence score
