N
NexusRoute
Back to Models

GPT-5.4 mini

OpenAImid

OpenAI's balanced mid-tier model that inherits many GPT-5.4 capabilities at roughly 70% reduced cost. Supports vision, tool use, and computer use with a 400K context window. The workhorse for production applications where quality and cost both matter.

Released 2026-03-05Knowledge cutoff: 2025-11
Medium confidence|Updated 58d ago|90% source confidence

Specifications

Context Window

400K tokens

Max Output

32K tokens

Input Price

$0.750 / 1M tokens

Output Price

$4.50 / 1M tokens

Latency Tier

Fast (speed score: 8.5/10)

Capability Profile

Structured Output
9/10
Instruction Following
9/10
Tool Use
9/10
Coding
8.5/10
Speed
8.5/10
Safety & Enterprise
8.5/10
Conversational
8.5/10
Reasoning
8/10
Long Context
8/10
Cost Efficiency
8/10
Factuality
8/10
Multimodal
7.5/10
Creativity
7.5/10

Feature Support

Vision Yes
Audio In No
Audio Out No
Video No
Image Generation No
Image Editing No
Function Calling Yes
JSON Mode Yes
Structured Output Yes
Streaming Yes
Reasoning No
Realtime No
Computer Use Yes
Web Search No

Best Use Cases

Production chatbots and assistant applications needing strong quality at manageable cost
Agentic tool-use workflows that don't need full GPT-5.4 intelligence
Document analysis and summarization within 400K context
Structured data extraction with reliable JSON mode
Computer use tasks with moderate complexity

Not Ideal For

The hardest reasoning problems — o3 or GPT-5.4 are noticeably better
Full multimodal pipelines needing audio/video (use GPT-5.4)
Contexts over 400K tokens
Tasks where the cheapest possible model is needed

Strengths

Excellent structured output compliance — near-perfect JSON schema adherence
Computer use capability at 70% lower cost than GPT-5.4
Fast inference with low TTFT for interactive applications
Strong instruction following inherited from the GPT-5 family

Weaknesses

No audio or video processing (text and images only)
Reasoning ceiling is measurably below GPT-5.4 on GPQA-level problems
Creative writing is competent but lacks the flair of the flagship
400K context is large but still 2.5x less than GPT-5.4

Edge Cases & Notes

Quality degrades gracefully past 300K tokens rather than cliff-dropping
Computer use is slightly less reliable than GPT-5.4 on multi-step GUI tasks
Batch API available at 50% discount making it very competitive for offline workloads

Provider Notes

The default recommendation for most production OpenAI deployments. Batch API makes it extremely cost-effective for async workflows. Available on Azure OpenAI Service.

Benchmarks

MMLU88.5%
HumanEval91.3%
Arena Elo1355

Benchmark Notes

MMLU-Pro 88.5%. Sits between Claude Sonnet 4.6 and GPT-5.4 on most benchmarks. SWE-bench Verified ~52%. Very strong cost-adjusted performance.

Research Meta

Last Evaluated

2026-04-01

Source Confidence

90%

Evaluation Method

LMSYS Arena, SWE-bench, MMLU-Pro, cost-quality Pareto analysis

Needs Re-evaluation

No

Sources

  • OpenAI GPT-5.4 mini announcement (Mar 2026)
  • LMSYS Chatbot Arena
  • Artificial Analysis quality-cost index

Continue exploring

Route a prompt

See how GPT-5.4 mini ranks

Compare models

Side-by-side analysis

Browse registry

Explore all 24 models