Model Details
GPT-5.4
Quick answer: GPT-5.4 is OpenAI's January 2026 frontier multimodal model, scoring 93.7% on ARC-AGI, 92.8% on GPQA Diamond, 82.7% on BrowseComp, 79.0% on CyberGym, and 39.8% on Humanity's Last Exam. It is the January 2026 point in GPT-5's incremental release cadence, superseded by GPT-5.5 in June 2026.
At a Glance
Where GPT-5.4 leads
- 93.7% ARC-AGI — strong abstract reasoning
- 92.8% GPQA Diamond — excellent graduate-level science
- 82.7% BrowseComp — strong web research
- 79.0% CyberGym — competitive cybersecurity
- Full multimodal capability
Where it lags
- Superseded by GPT-5.5 on most benchmarks
- 39.8% HLE — below frontier models on Humanity's Last Exam
- Proprietary
Best for: Production use cases validated on GPT-5.4; users migrating from GPT-5.x to this specific version.
What GPT-5.4 Is
GPT-5.4 (released January 2026) is one of OpenAI's incremental releases in the GPT-5 family. Following the GPT-4 family's cadence (gpt-4, gpt-4-turbo, gpt-4o), GPT-5 has been released as a versioned series with incremental improvements.
GPT-5.4's 93.7% ARC-AGI and 92.8% GPQA Diamond represent strong frontier performance. It was superseded by GPT-5.5 (June 2026) which improves on most benchmarks including ARC-AGI (95.0% vs 93.7%) and HLE (41.4% vs 39.8%).
Specifications
| Field | Value |
|---|---|
| Organization | OpenAI |
| License | Proprietary (API only) |
| Release date | January 2026 |
| Knowledge cutoff | September 2025 |
| Modality | Multimodal |
Pricing
Available via OpenAI API. Refer to the OpenAI pricing page.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| ARC-AGI | 93.7% | Benchgen evaluation | 2026-01 |
| GPQA Diamond | 92.8% | Benchgen evaluation | 2026-01 |
| BrowseComp | 82.7% | Benchgen evaluation | 2026-01 |
| CyberGym | 79.0% | Benchgen evaluation | 2026-01 |
| Humanity's Last Exam | 39.8% | Benchgen evaluation | 2026-01 |
| Agents Last Exam | 37.3% | Benchgen evaluation | 2026-01 |
GPT-5.4 vs Alternatives
| Model | ARC-AGI | GPQA Diamond | HLE | Version |
|---|---|---|---|---|
| GPT-5.4 | 93.7% | 92.8% | 39.8% | Jan 2026 |
| GPT-5.5 | 95.0% | 93.6% | 41.4% | Jun 2026 |
| Gemini 3.1 Pro | 98.0% | 94.3% | 46.4% | Mar 2026 |
GPT-5.5 strictly dominates GPT-5.4 on all available benchmarks. For new deployments, GPT-5.5 is the recommended GPT-5 series model.
Related Models & Benchmarks
- Compare: GPT-5.5 · Gemini 3.1 Pro · GPT-5 Codex
- Benchmarks: ARC-AGI · GPQA Diamond · Humanity's Last Exam
Frequently Asked Questions
What is GPT-5.4?
OpenAI's January 2026 frontier multimodal model scoring 93.7% ARC-AGI, 92.8% GPQA Diamond, and 39.8% HLE. Superseded by GPT-5.5 (June 2026).Should I use GPT-5.4 or GPT-5.5?
GPT-5.5 for new deployments — strictly better on all benchmarks (95.0% vs 93.7% ARC-AGI, 41.4% vs 39.8% HLE).Specs from OpenAI's GPT-5.4 release (January 2026) and Benchgen evaluations. Last updated 2026-07-24.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.