GPT-5.5 in Codex and Goal Mode
OpenAI's newest frontier model lands in Codex as its strongest agentic coding model yet — alongside Goal Mode, Appshots, and remote computer use.
GPT-5.5: the strongest agentic coding model
OpenAI's newest frontier model for coding, computer use, and knowledge work — recommended for most Codex tasks.
82.7%
Terminal-Bench 2.0 (SOTA)
58.6%
SWE-Bench Pro
#1
Strongest agentic coding model to date
GPT-5.5 excels at implementation, refactors, debugging, testing, and validation, and reaches further across tools until a task is finished. In Codex it delivers better results with fewer tokens than GPT-5.4 for most users.
- Best for implementation, refactors, debugging, testing & validation
- Fewer tokens than GPT-5.4 for most users
- Recommended default when it appears in your Codex model picker
Codex Goal Mode
Set an objective and let Codex drive toward it — now standard across the app, IDE extension, and CLI.
Goal Mode lets you hand Codex a specific objective and have it work toward that goal for hours or even days, instead of step-by-step instructions. It moved from experimental to standard in May 2026 and is enabled by default in Codex CLI v0.133.0.
- •Drives toward a defined objective over long, multi-step runs
- •Available across the Codex app, IDE extension, and CLI
# Codex CLI - describe the goal, not the steps
codex "Goal: make the test suite
pass on Node 22 and open a PR"More from Codex in May 2026
GPT-5.5 and Goal Mode shipped alongside a batch of agentic upgrades.
Appshots (macOS)
Press both Command keys to send the frontmost app window — screenshot plus available text — to Codex, with no copy, paste, or manual description.
Remote computer use
Codex can keep using desktop apps after your Mac locks, including remotely via Codex Mobile, with short-lived authorization, covered displays, and relock on local input.
Advanced browser annotations
Tweak styling — font size, colors, spacing — directly with annotations in the in-app browser to give Codex a clearer signal while editing.
Faster browser & CLI updates
The in-app browser extracts image assets and structured data faster; the CLI added local search, permission profiles, and MCP improvements through v0.135.0.
GPT-5.5 vs Claude Opus 4.8
Both are top-tier in May 2026. A quick, honest read — not a verdict.
| Aspect | GPT-5.5 (Codex) | Claude Opus 4.8 |
|---|---|---|
| Published coding benchmark | Terminal-Bench 2.0 82.7%, SWE-Bench Pro 58.6% | Online-Mind2Web 84%; no SWE-bench score published |
| Signature agentic feature | Goal Mode — drive toward an objective for hours/days | Dynamic workflows — orchestrate up to 1,000 subagents |
| Reliability claim | Better results with fewer tokens than GPT-5.4 | ~4× less likely to let code defects pass undetected |
Numbers are each vendor's own published figures and aren't directly comparable. Full GPT-5.5 vs Claude 4.8 comparison →
One QCode account for both Claude and Codex
QCode gives you access to both Claude (Claude Code) and OpenAI Codex / GPT through a single account, optimized for low latency in China — so you can run GPT-5.5 Goal Mode and Claude Code workflows without juggling two foreign subscriptions, cards, or phone numbers.
Use GPT-5.5 and Codex through QCode
Get Codex and Claude access from one China-optimized account — sign up and start in minutes.