GPT-5: Features, Pricing, and Coding Use
GPT-5 is an OpenAI reasoning model for coding, agentic tasks, and general problem solving. The API model supports configurable reasoning effort, text and image input, a 400,000-token context window, and up to 128,000 output tokens. OpenAI now labels it a previous model and recommends a newer generation for current builds.
What Is GPT-5
OpenAI released the model in the API alongside mini and nano variants. The main gpt-5 endpoint emphasized reasoning, coding, instruction following, and tool use. It also introduced controls such as minimal reasoning effort and response verbosity.
It is a hosted proprietary model, not an open-weight release. Access depends on the API, a product that includes the model, or a supported third-party integration.
GPT-5 Product and API Model Distinction
It is inaccurate to describe the model as a literal combination of GPT-4o, o3, and Codex. At launch, the ChatGPT product used a routed system that could choose between faster and deeper reasoning behavior. The API's gpt-5 ID referred to a specific reasoning model, while a separate chat-oriented ID served the non-reasoning ChatGPT model.
Codex is also a coding product and model line, not a component that should be described as merged into every response. Use the exact product surface and model ID when documenting behavior.
GPT-5 vs Claude Sonnet 4.6 vs Gemini 3 Pro
A useful comparison starts with workload shape rather than a broad winner.
| Workload question | GPT-5 | Claude Sonnet 4.6 | Gemini 3 Pro |
|---|---|---|---|
| Reasoning control | Multiple reasoning-effort levels | Adaptive and extended reasoning controls | Thinking-level controls |
| Long input | 400K context | Model-specific current context | Model-specific current context |
| Coding route | API and OpenAI coding surfaces | Claude API and Claude Code | Gemini API and Google coding surfaces |
| Current status | Previous generation | Check current Claude catalog | Check current Gemini catalog |
Run the same task with the same repository snapshot, tools, time budget, and acceptance tests. Provider benchmark tables are not interchangeable when their harnesses differ.
Coding & SWE-bench
OpenAI reported 74.9% on SWE-bench Verified at launch, with 23 of 500 problems omitted because their solutions did not reliably pass on OpenAI's infrastructure. The model also received a prompt that emphasized verification. Keep those conditions with the number.
For your own evaluation, try:
- “Fix this issue and show the failing test before the patch and the passing test after it.”
- “Refactor this interface without changing its public behavior or dependency direction.”
- “Plan a migration, identify rollback points, and stop before editing files.”
API Pricing and Access
OpenAI's current model page lists $1.25 per million input tokens, $0.125 per million cached input tokens, and $10 per million output tokens. Pricing for tools can be separate. The page also identifies this release as a previous model, so new implementations should compare it with the recommended current generation before locking an endpoint.
The 400K context window is not a promise that every prompt should fill it. Long inputs raise latency, review burden, and cost, and irrelevant context can reduce task focus.
Using GPT-5 in Verdent
Verdent's current built-in catalog lists newer generations rather than the original model. Check Verdent model pricing for enabled IDs. Verdent also documents a BYOA route for Codex, but that connects an agent runtime and does not prove that it runs this original model.
Compare Reasoning Settings on One Task
Use a fixed repository task and test minimal, low, and higher reasoning settings. Measure success, latency, token use, and review time together.
Compare current GPT IDs in Verdent Review the Codex agent route in Verdent
Frequently Asked Questions
What is this model's API context window?
The API model supports a 400,000-token context window and up to 128,000 output tokens. Practical limits also depend on input mix, tool use, and account constraints.
Is this model a combination of GPT-4o, o3, and Codex?
No. That description conflates a routed ChatGPT product system, model families, and the Codex coding product. The API ID refers to a specific reasoning model.
What does this API model cost?
OpenAI currently lists $1.25 per million input tokens, $0.125 per million cached input tokens, and $10 per million output tokens. Tool charges can be separate.
Is the original release built into Verdent?
Verdent's current catalog lists newer models, not the original ID. Confirm the enabled model before starting a task.