AI

GPT-5.5

SOTAFoundation

OpenAI's smartest frontier model for complex professional work, with a 1M token context window and natively omnimodal architecture.

Model Overview

GPT-5.5 is OpenAI's most capable model as of April 2026. It processes text, images, audio, and video in a single unified architecture — unlike previous multimodal models that stitched separate models together. It excels at agentic coding, computer use, knowledge work, and scientific research. The model supports a 1M token context window in the API and 400K in Codex. It was co-designed with NVIDIA GB200/GB300 hardware and used Codex to rewrite its own serving infrastructure before launch, improving token generation speeds by over 20%.

Capabilities

Inputs

Text, image, audio, video

Outputs

Text, code

Task categories

Reasoning, coding, research, data analysis, agentic tasks, long-context understanding

Languages

Multilingual

Tags

Reasoning, agentic, multimodal, long-context, coding, omnimodal

Model Details

Developer
OpenAI
Version
gpt-5.5-2026-04-23
Release date
April 23, 2026
Model type
Foundation
Context window
1M tokens
Open source
No
Commercial use
Allowed
Fine-tunable
No
License
OpenAI proprietary

Architecture

Type
Transformer (omnimodal unified architecture)
Base model
GPT-5.4
Active params
Not officially disclosed
Total params
9.7 trillion

Benchmarks

BixBench

80.5%

ARC AGI 2

85%

GeneBench

25%

Expert SWE

73.1%

Terminal Bench

82.7%

Safety & Compliance

Notes
Additional safeguards described in updated system card released April 24, 2026
Red teamed
Yes
Training
RLHF, reinforcement learning, supervised fine-tuning
Fine-tunable
No
Base model
GPT-5.4
Usage recommendations
TBA