Stronger agent coding
OpenAI reports higher completion on agent coding evals such as Terminal-Bench—better suited to multi-step coding and tool collaboration.
OpenAI · GPT-5.5 · April 2026
GPT-5.5 — general-purpose flagship for coding, analysis, and tool use
- Million-token context
- Extra-long single replies
- Tunable depth tiers
- Agent-coding workhorse generation
What is GPT-5.5?
GPT-5.5 is OpenAI’s flagship released in April 2026 for complex professional work: agent coding, computer use, knowledge work, and early research. Versus GPT-5.4, OpenAI emphasizes stronger agent capability and lower token use. With GPT-5.6 available, the heaviest new jobs can compare Sol and everyday balanced work can compare Terra; if your workflows already run stably on GPT-5.5, it remains a solid primary. On iMini Agent, pick GPT-5.5 to use it.
- Vendor
- OpenAI
- Released
- April 2026
- Context
- 1M tokens
- Max output
- 128K tokens
- Deep reasoning
- Tunable tiers
- Best as
- Flagship primary for validated workflows
What’s new versus GPT-5.4
Agent coding, intelligence at matched latency, and knowledge work—what to check before moving from 5.4 to 5.5.
More intelligence at similar latency
Versus GPT-5.4, OpenAI emphasizes stronger intelligence at comparable serving latency, and finishing Codex-style work with fewer tokens.
Steadier long-context professional work
Million-token context keeps long specs, multi-source materials, and tool traces in one thread—suited to knowledge work and complex delivery.
Higher research and security eval ladder
On GeneBench and cybersecurity-related evals, OpenAI reports step-ups versus the prior generation—useful before deciding whether to move to GPT-5.6.
Official evaluations
Results OpenAI published with GPT-5.5, compared with GPT-5.4 and peer flagships—covering composite intelligence, software engineering, knowledge work, computer use, and security research.
Three common workflows
GPT-5.5 still fits validated complex professional work; new heaviest loads can compare GPT-5.6.
Agent coding & repo changes
For multi-step coding, test repair, and cross-file features. Teams whose prompts and evals already bind to GPT-5.5 can keep shipping on it.
Knowledge work & decision memos
For research synthesis, table work, and one-page decision memos. Quote hard constraints and give a checkable recommendation.
Tool and computer-use flows
For workflows that call tools repeatedly and finish end to end. Pin success checks and abort conditions up front.
How to choose vs GPT-5.6 Sol and GPT-5.6 Terra
All three are OpenAI workhorses. The differences are generation depth, everyday cost, and whether your workflow is already validated.
| Dimension | GPT-5.5 | GPT-5.6 Sol | GPT-5.6 Terra |
|---|---|---|---|
| Context | 1M tokens | 1M tokens | 1M tokens |
| Reasoning | Tunable depth tiers | Max depth · ultra available | Balanced depth |
| Coding / agents | Flagship coding & agents | Heaviest current long-horizon coding | Day-to-day coding & mid-weight agents |
| Long tool runs | Validated workflows can continue | Flagship depth + ultra | More balanced completion and throughput |
| Speed posture | Professional-work primary | Completion quality and depth first | Everyday primary favors balance |
| Prefer when | Workflows already validated on 5.5 | New heaviest reasoning and long-horizon jobs | Near–5.5 quality on the GPT-5.6 balanced tier |
Using GPT-5.5 on iMini
Free quota, model naming, key specs, and how it differs from GPT-5.6.
Use GPT-5.5 free on iMini Agent
Million-token context, tunable depth, suited to validated professional workflows.
Open GPT-5.5No OpenAI API key required.