How to access and use GPT-5 [2026]

GPT-5 was a major step forward in OpenAI’s reasoning and coding models when it launched in August 2025.

But here's the thing: most people don't know how to access it properly or use it effectively.

Here’s how GPT-5 access works in 2026, where the model is still available, and how to use it effectively for the workloads it still supports.

Current status: GPT-5 is now an earlier OpenAI model generation. OpenAI’s current frontier family is GPT-5.6, led by GPT-5.6 Sol. GPT-5 remains relevant for existing API workflows and for comparing how OpenAI’s reasoning models have evolved.

What you'll learn:

• Free vs. paid access options (and which to choose)
• Advanced features that save hours of work
• Real-world prompting techniques that get better results
• Cost optimization strategies for heavy users

Let's get started.

What is GPT-5?

GPT-5 is an earlier OpenAI reasoning model built for coding, reasoning, and agentic tasks, with configurable reasoning effort.

Released on August 7, 2025, GPT-5 launched with three modes for different types of work.

At launch, GPT-5 was presented in three modes:

• Auto mode - delivers fast responses for everyday tasks
• Thinking mode - provides deeper analysis for complex problems
• Pro mode - offers extended reasoning for professional-level work

Key performance improvements over GPT-4o:

• Strong coding performance - GPT-5 scored 74.9% on SWE-bench Verified and 88% on Aider Polyglot in OpenAI’s launch evaluations.
• Strong mathematics performance - GPT-5 scored 94.6% on AIME 2025 without tools in OpenAI’s launch evaluation.
• Strong scientific reasoning - GPT-5 Pro scored 88.4% on GPQA without tools in OpenAI’s launch evaluation.
• More efficient coding - On SWE-bench Verified, GPT-5 used 22% fewer output tokens and 45% fewer tool calls than o3 at high reasoning effort, while scoring 74.9%.

How to Access GPT-5

GPT-5 is no longer available in ChatGPT, but it remains available through the OpenAI API. Here’s how GPT-5 access works in 2026:

Free Access

GPT-5 is not currently available for free in ChatGPT. OpenAI retired GPT-5 Instant and Thinking from ChatGPT in February 2026.

If you still need GPT-5 specifically, the model remains available through the OpenAI API, where usage is billed by tokens. For new workloads, OpenAI now recommends starting with GPT-5.6 instead.

ChatGPT Plus ($20/month)

ChatGPT Plus still costs $20 per month, but it no longer provides access to GPT-5 itself because GPT-5 was retired from ChatGPT in February 2026.

ChatGPT Pro ($200/month)

ChatGPT’s $200 Pro tier remains the highest-usage Pro option. OpenAI also offers a $100 Pro tier with the same core capabilities but lower usage allowances.

OpenAI also offers ChatGPT Business and ChatGPT Enterprise for teams that need shared workspaces, admin controls, and higher organizational limits.

Other Ways to Access GPT-5

Microsoft 365 Copilot

Microsoft added GPT-5 to Microsoft 365 Copilot in August 2025. Since then, Copilot has moved to newer OpenAI models, with GPT-5.6 now available across Word, Excel, PowerPoint, Chat, and Cowork.

If you’re using Microsoft 365 Copilot today, you’re more likely to encounter the newer GPT-5.6 family than the original GPT-5 model.

GitHub Copilot

GitHub Copilot added GPT-5 in 2025, but its current model picker has moved to newer OpenAI models. GitHub now supports GPT-5.6 Sol, Terra, and Luna across Copilot, depending on the plan and feature.

If you’re using Copilot today, choose the model based on the coding task rather than looking specifically for the original GPT-5.

API Access

GPT-5 remains available through the OpenAI API even though it has been retired from ChatGPT. For setup steps, authentication, and request examples, see our ChatGPT API guide.

For developers maintaining an existing GPT-5 workflow:

• Sign in at platform.openai.com
• Create or manage your API key
• Use GPT-5 through supported endpoints such as Chat Completions
• Pay based on token usage and the GPT-5 variant you select

OpenAI now recommends the GPT-5.6 family for most new workloads, so GPT-5 makes more sense here when you are maintaining or evaluating an existing implementation rather than starting from scratch.

Chatbase (No-Code GPT-5 Training)

You can train an AI model on your own data in Chatbase without building the retrieval and deployment setup from scratch.

Chatbase’s current model catalog includes GPT-5 Mini, GPT-5 Nano, and newer GPT-5-generation models such as GPT-5.5, GPT-5.6 Terra, and GPT-5.6 Luna.

• Add business knowledge from sources such as your website, documents, and connected content
• Set instructions for how the agent should respond and handle different requests
• Connect approved actions and integrations when the workflow needs them
• Deploy the agent for AI customer support and hand conversations to a person when needed

Chatbase has a Free plan. For current paid-plan limits and pricing, check the Chatbase pricing page

GPT-5 API Models Comparison

OpenAI offers three GPT-5 variants through the API, each optimized for different use cases and budgets:

GPT-5

• Best for: Complex reasoning, coding, and professional applications
• Cost: Input $1.25, Output $10.00 (per 1M tokens)
• Purpose: Built for coding, reasoning, and agentic tasks across domains, with configurable reasoning effort from minimal to high.
• Context Window: 400,000 tokens
• Max Output: 128,000 tokens
• Knowledge Cutoff: September 30, 2024

GPT-5-mini

• Best for: General chatbots, content generation, and moderate complexity tasks
• Cost: Input $0.25, Output $2.00 (per 1M tokens)
• Purpose: Cost-effective variant for general tasks with similar capabilities to GPT-5 but optimized for speed and efficiency; supports reasoning levels (minimal to High), suitable for lighter applications while maintaining low hallucinations and good instruction adherence
• Context Window: 400,000 tokens
• Max Output: 128,000 tokens
• Knowledge Cutoff: May 31, 2024

GPT-5-nano

• Best for: High-volume, simple tasks like classification or basic Q&A
• Cost: Input $0.05, Output $0.40 (per 1M tokens)
• Purpose: Ultra-efficient, low-cost model for simple tasks and high-volume usage; offers basic reasoning capabilities with the same safety features and optimizations as larger siblings
• Context Window: 400,000 tokens
• Max Output: 128,000 tokens
• Knowledge Cutoff: May 31, 2024

What GPT-5 is good at

Reasoning and Problem-Solving: At launch, ChatGPT’s GPT-5 system used a router to decide when to answer quickly and when to use deeper reasoning. In the API, developers can control reasoning effort directly, from minimal to high, depending on the task.

Coding and Development: In OpenAI’s internal testing, GPT-5 beat o3 on frontend web development 70% of the time. It was also designed to work across larger codebases, fix bugs, edit code, and follow detailed development instructions.

Speed and Efficiency

• Coding efficiency: On SWE-bench Verified, GPT-5 used 22% fewer output tokens and 45% fewer tool calls than o3 at high reasoning effort.
• Context window: GPT-5 supports a 400,000-token context window in the API.
• API pricing: GPT-5 costs $1.25 per million input tokens and $10 per million output tokens.

Multimodal Capabilities: GPT-5 can reason over images and other non-text inputs. OpenAI’s launch evaluations included visual, video-based, spatial, and scientific reasoning tasks.

Where GPT-5 isn't good at

Real-World Orchestration: GPT-5 scored 43.72% on MCP-Universe, a benchmark built around realistic multi-step tasks using external tools. The result shows that long workflows and unfamiliar tool environments remained difficult for the model.

Creativity and Tone: Style quality depends heavily on the prompt and the type of writing. GPT-5 can follow detailed tone and formatting instructions, but subjective preferences such as humor, warmth, or conversational style can still vary by user and task.

Prompt Sensitivity: GPT-5 responds best to clear, consistent instructions. Vague or conflicting prompts can reduce reliability, especially in longer workflows or tool-heavy tasks.

Simple, High-Volume Tasks: Full GPT-5 can be unnecessary for straightforward workloads. OpenAI positions GPT-5 mini as a faster, more cost-efficient option for well-defined tasks, while GPT-5 nano is designed for simple, high-volume work such as summarization and classification.

How to Build an AI Agent with GPT-5 (with Chatbase)

Building an AI agent goes beyond simple chatbots. You're creating an autonomous system that can perform tasks, make decisions, and integrate with external tools, like booking appointments, processing orders, or managing customer support.

Chatbase lets you build AI agents using your own business knowledge, instructions, and connected actions. Chatbase supports several GPT-5-generation models, including GPT-5 Mini, GPT-5 Nano, and newer GPT-5.x options, so you can choose a model based on the agent’s workload.

What You'll Need

• Chatbase Account
• Create a free Chatbase account to build and test your agent.
• Paid plans add higher message limits, advanced models, integrations, analytics, and other features. Check Chatbase pricing for current limits and costs.

Data Sources

• FAQs, PDFs, website content, or Notion pages for training
• Common questions and business policies

Optional Integrations

• Zapier, Slack, Calendly, or custom APIs for advanced automation