GPT-5 Codex
GPT-5 Codex is a coding-oriented variant of OpenAI's GPT-5 model line, tuned for software engineering tasks such as code generation, refactoring, and agentic development workflows.
API Pricing
| Provider | Input / 1M | Output / 1M |
|---|---|---|
| $1.25 | $10.01 |
Prices updated daily. Last check: Nov 20, 2026
Compare API pricing for every OpenAI model →GPT-5 Codex pricing by provider
Input, output, and batch rates, plus alternatives, for one provider at a time.
Performance & Benchmarks
Source: Artificial Analysis →Benchmarks measured Nov 2026. Scores are independent evaluations, not vendor-reported.
Model Details
General
- Creator
- OpenAI
- Modalities
- Text
Capabilities
- Open Source
- No
Strengths & Limitations
Strengths
- Tuned specifically for software engineering tasks rather than general-purpose chat
- Part of the GPT-5 family, so it inherits that generation's training and alignment work
- Designed for long-running agentic coding sessions that span many turns and files
- Exposed through the standard OpenAI chat-completions interface, easing integration with existing clients
- Associated with OpenAI's Codex surfaces (CLI, IDE extensions, cloud agents), giving ready-made tooling paths
- Suited to multi-file repository work such as refactors, migrations, and test authoring
Limitations
- Specialization toward coding makes it a less natural fit for creative writing, marketing, or general conversational workloads
- Context window is not tracked in our database — confirm limits with OpenAI documentation before designing long-repository prompts
- Model weights are served via API only; self-hosting is not an option
- Throughput and time-to-first-token figures in our benchmark record are unpopulated, so latency expectations should be validated against your own workload
- Provider availability for Codex-branded variants is narrower than for general GPT-5 models
Key Features
About GPT-5 Codex
Common Use Cases
GPT-5 Codex fits development workflows where the model is expected to act on a codebase rather than just discuss it: implementing a feature from a ticket description, performing framework or language migrations across many files, generating unit and integration tests, and reviewing pull request diffs for defects. Its orientation toward agentic operation makes it a candidate for autonomous or semi-autonomous coding agents that plan, edit, run, and iterate over multiple turns, as well as for IDE assistants that need to reason about surrounding project context. Teams running high-volume, non-coding text workloads — summarization, classification, customer-facing chat — will usually get better cost and behavior alignment from a general-purpose or lightweight model instead, reserving Codex for engineering tasks where code quality dominates.
Frequently Asked Questions
How much does GPT-5 Codex cost?
Pricing varies by provider and by pricing type — input tokens, output tokens, cached input, and any batch or committed-use discounts are all billed differently. Rates also change frequently. See the pricing table on this page for the current figures from each provider that serves this model.
What is GPT-5 Codex best used for?
Software engineering work: generating and editing code across a repository, refactoring, writing tests, debugging, reviewing diffs, and powering coding agents that run for many turns on a single task. It is specialized for these workloads rather than general conversation or content writing.
How does GPT-5 Codex differ from a general GPT-5 model?
Both come from OpenAI's GPT-5 family, but the Codex variant is tuned around coding and agentic development behavior. General GPT-5 models are broader and better suited to mixed workloads, while GPT-5 Codex concentrates on repository-level engineering tasks.
What context window does GPT-5 Codex support?
We do not currently track a confirmed context window for this entry. Because agentic coding sessions can consume large amounts of context, verify the current limit in OpenAI's model documentation or with your chosen provider before relying on very long repository prompts.
Can I run GPT-5 Codex on my own hardware?
No — it is served through API providers rather than distributed as downloadable weights. If self-hosting is a requirement, you would need to compare against open-weight coding models instead.