New Models Available

OpenAI has released two new models in the GPT-5.4 family:

• GPT-5.4 mini: A faster, more efficient mid-tier model optimized for coding, reasoning, and tool use
• GPT-5.4 nano: The smallest and cheapest version, designed for classification, data extraction, and simple tasks

Performance & Capabilities

GPT-5.4 mini significantly upgrades the previous GPT-5 mini:

• Achieves 54.4% on SWE-Bench Pro (vs. 45.7% for GPT-5 mini) with 2x faster latency
• Approaches full GPT-5.4 performance on several coding benchmarks
• Strong multimodal capabilities for computer use tasks (72.1% on OSWorld-Verified)
• Handles tool calling, function use, web search, file search, and computer vision

GPT-5.4 nano is ideal for high-volume, latency-sensitive tasks:

• Achieves 52.4% on SWE-Bench Pro despite being the smallest variant
• Optimized for simple coding tasks, classifications, and supporting agents
• Significantly outperforms GPT-5 nano across all benchmarks

Availability & Pricing

API Access:

• GPT-5.4 mini: $0.75 per 1M input tokens, $4.50 per 1M output tokens; supports 400k context window, images, tools, web search, computer use, and skills
• GPT-5.4 nano: $0.20 per 1M input tokens, $1.25 per 1M output tokens (API only)

Integration Points:

• Both models available in the OpenAI API
• GPT-5.4 mini available in Codex (consuming 30% of GPT-5.4 quota) and ChatGPT
• Mini model supports subagent delegation in Codex for distributed task execution

Use Cases

These models are built for workloads where latency and cost directly impact user experience: coding assistants needing real-time responsiveness, subagent systems handling parallel tasks, and computer vision applications processing screenshots. The models excel in architectural patterns where larger models handle planning while smaller models execute narrower subtasks at scale.