GPT-5.3 Codex
GPT-5.3 Codex is OpenAI's flagship coding model. It leads Terminal-Bench by a wide margin and performs on par with Opus 4.6 on our internal benchmarks, at roughly one-third the price. Much faster than previous GPT generations. A strong default for most coding tasks.
Strengths
- Leads Terminal-Bench by a wide margin. Competitive with Opus 4.6 on our internal benchmarks.
- About one-third the price of Opus with comparable quality on most tasks. Good for daily coding, long debugging sessions, and cost-conscious teams.
- Grinds through complex, multi-step problems and deep debugging sessions.
Limitations
- Code style is less polished than Opus on architecture-heavy tasks.
- Terminal-Bench skews toward general reasoning; real-world coding gains may differ.
Tools
GPT-5.3 Codex has access to all agent tools when used with Cursor including:
Find files by name, read directory structures, and grep for patterns.
Generate search queries and fetch results from the web.
Read file contents, including images for vision-capable models.
Suggest edits and apply them automatically.
Execute terminal commands and monitor output.
Control a browser to take screenshots, test applications, and verify visual changes. See the Browser documentation.
Generate images from text descriptions or reference images.
Ask clarifying questions while continuing to work in the background.
Retrieve rules based on type and description.
Learn more about how tools work and tool calling fundamentals.
Pricing
Cursor plans include two usage pools. GPT-5.3 Codex draws from the third-party Other Models pool, which charges at the rates below. Individual plans include at least $20 of third-party Other Models usage each month (more on higher tiers). All prices are per million tokens.
| Name | ||||
|---|---|---|---|---|
| $1.75 | - | $0.175 | $14 |
A high reasoning effort variant (gpt-5.3-codex-high) is available for tasks that need deeper analysis.