GPT-5.4, the OpenAI model in numbers

Overview

OpenAI released GPT-5.4 on March 5, 2026. It has a reasoning mode: it can "think" before answering, which helps with multi-step problems but uses more tokens, and therefore more money. Its weights are not published, so it only runs remotely, through the OpenAI API or partner hosts (48 listed on models.dev). That makes it 7 months old as of October 22, 2026.

We file it under frontier models (24 models in this category on the site).

Benchmark results

According to results published by Epoch AI (checked on October 22, 2026), GPT-5.4 has a capability index of 87.2 out of 100, computed on 6 benchmarks from our basket.

That puts it at #13 in our weekly model ranking.

• GPQA Diamond (PhD-level science questions): 93.3% — the best tracked model, Gemini 4 Argon, reaches 96.4%
• OTIS Mock AIME (competition mathematics problems): 97.8% — the best tracked model, GPT-5.5 Pro, reaches 100%
• FrontierMath (very hard research-level mathematics): 78.6% — the best tracked model, GPT-6 Astra, reaches 93.7%
• SWE-bench Verified (fixing real software bugs): 76.9% — the best tracked model, Claude Opus 4.7, reaches 83.5%

What it can do

Its context window reaches 1,050,000 tokens, roughly 1,575 pages of text at 500 words per page: enough to load a whole book or a mid-sized codebase. A single answer can run to 128,000 tokens.

It accepts as input: text, image and PDF. Its knowledge stops at August 31, 2025: beyond that, it needs web search or your documents.

It can call tools (functions, APIs) and return structured JSON output, which makes it a possible foundation for an AI agent.

Pricing and budget

GPT-5.4 costs $2.5 per million input tokens and $15 per million output tokens (median of 48 providers, checked on October 22, 2026).

In practice, an agent task that reads 20,000 tokens and writes 2,000 costs about $0.08, or $80 for 1,000 such tasks.

That is in line with frontier models, whose median output price is $14.5.

To estimate your own budget, use the agent cost calculator.

What is it good for?

Given its category, GPT-5.4 is mainly aimed at hard tasks where quality matters more than cost: deep reasoning, complex code, long analyses.

Its SWE-bench Verified result (76.9%, above the median of tracked models) makes it a serious candidate for coding agents.

Its very large context window makes it worth considering for analysing large sets of documents in a single request.

Before adopting it, test it on your own examples: benchmarks measure standard tasks, not your job.

What could replace it?

In the same category, the models closest to GPT-5.4 are:

• GPT-5.6 Sol (OpenAI, index 87.3)
• Gemini 3 Pro Preview (Google DeepMind, index 87)
• Qwen3.7 Max (Alibaba Cloud (Qwen), index 86.6)
• Grok 4.6 (xAI, index 86.1)