AI/TLDR

GPT-5

OpenAI's unified flagship that routes between a fast model and deep reasoning.

Overview

GPT-5 is OpenAI's flagship large language model, released on August 7, 2025. Rather than shipping as a single model, GPT-5 is a unified system: a fast, high-throughput model (gpt-5-main) handles most queries, a deeper reasoning model (gpt-5-thinking) tackles harder problems, and a real-time router decides which to use based on the conversation's complexity, tool needs, and your explicit intent. This let OpenAI fold its previous fast and reasoning model families into one product.

Available through the OpenAI API and ChatGPT, GPT-5 supports text and image input with text output, a 400,000-token context window, and up to 128,000 output tokens (which include the model's internal reasoning tokens). Its training data has a knowledge cutoff of September 30, 2024. A higher-effort variant, GPT-5 pro, offers extended reasoning for the most demanding tasks.

At launch GPT-5 posted state-of-the-art results on math, coding, and science benchmarks while cutting hallucination rates relative to earlier OpenAI models, and it priced aggressively at $1.25 per million input tokens and $10 per million output tokens. It has since been superseded by GPT-5.1 and later releases in the GPT-5 line, with OpenAI steering new development toward the newer variants.

Released2025-08-07
LicenseProprietary
WeightsAPI only
ParametersNot disclosed
Context400K
Max output128K
ArchitectureUnified system combining a fast, high-throughput model (gpt-5-main) and a deeper reasoning model (gpt-5-thinking), with a real-time router that decides which to use based on conversation type, complexity, tool needs, and explicit intent. Trained in three stages: unsupervised pretraining, supervised fine-tuning, and reinforcement learning. Exact parameter count is not disclosed.
Knowledge cutoffSeptember 30, 2024
ModalitiesText, Vision
StatusSuperseded

Benchmarks

  1. AIME 2025 (no tools)94.6%
  2. SWE-bench Verified74.9%
  3. GPQA Diamond (no tools)88.4%
  4. Aider Polyglot88%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

Pricing

Input$1.25 / 1M tokens per 1M tokens
Cached input$0.125 / 1M tokens per 1M tokens
Output$10.00 / 1M tokens per 1M tokens

Launch API pricing for the full GPT-5 model; mini and nano variants are cheaper. Cached input receives a 90% discount.

Pricing source ↗

Strengths

  • Unified router automatically picks fast answers or deep reasoning, removing the need to manually choose a model
  • Strong competition-math performance (94.6% on AIME 2025 without tools)
  • Leading agentic coding results (74.9% on SWE-bench Verified)
  • Large 400K-token context window for full codebases and long documents
  • Lower hallucination and error rates than prior OpenAI models
  • Aggressive API pricing ($1.25/$10 per million tokens) with a 90% cached-input discount

Best for

  • Agentic coding: multi-step bug fixing, refactoring, and pull-request-scale changes across a repository
  • Complex reasoning over math, science, and finance problems
  • Long-document and full-codebase analysis using the 400K context window
  • Tool-using agents that chain API calls and external actions
  • Drafting, editing, and technical writing with reduced hallucination
  • Multimodal tasks that combine text prompts with image input

How to access

ProviderModel ID
OpenAI ↗gpt-5

GPT (Flagship / Thinking) — every version

The full lineage of the GPT (Flagship / Thinking) line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.

VersionReleasedContextLicense
GPT-5.5current2026-04-231.05MProprietary
GPT-5.42026-03-05Proprietary
GPT-5.22025-12-11Proprietary
GPT-5.12025-11-12Proprietary
GPT-52025-08-07Proprietary
GPT-4o2024-05-13Proprietary

FAQ

When was GPT-5 released?

GPT-5 was released on August 7, 2025, available through ChatGPT and the OpenAI API. It has since been superseded by GPT-5.1 and later models in the GPT-5 line.

What is GPT-5's context window and max output?

GPT-5 has a 400,000-token context window and can produce up to 128,000 output tokens, which include the model's internal reasoning tokens.

How much does the GPT-5 API cost?

At launch the full GPT-5 model was priced at $1.25 per million input tokens and $10 per million output tokens, with cached input at $0.125 per million (a 90% discount). Mini and nano variants cost less.

What makes GPT-5 different from earlier OpenAI models?

GPT-5 is a unified system, not a single model. A real-time router decides whether to answer with a fast model (gpt-5-main) or a deeper reasoning model (gpt-5-thinking) based on the query's complexity, so you no longer have to pick a model manually.