Encyclopedia

GPT-5 is a large language model developed by OpenAI and released on August 7, 2025. It serves as the successor to GPT-4 and represents what OpenAI describes as its "best AI system yet," featuring significant improvements in intelligence, reasoning capabilities, and performance across various domains including coding, mathematics, writing, health, and visual perception.¹ The model powers the latest version of ChatGPT and is designed as a unified system that can dynamically adjust its computational resources based on query complexity.

Overview

GPT-5 is structured as a unified system comprising multiple components: a smart, efficient model for handling most queries, a deeper reasoning model (GPT-5 thinking) for complex problems, and a real-time router that determines which model to use based on conversation type, complexity, tool requirements, and explicit user intent.¹ The routing system is continuously trained on real signals, including user model-switching behavior, preference rates for responses, and measured correctness metrics.

The model represents a significant advancement in OpenAI's capabilities, with CEO Sam Altman claiming it performs at the level of experts in various scientific fields.² However, the launch was met with mixed reactions, particularly regarding changes to the model's conversational personality that users found less warm and supportive than previous versions.²³

Development and Release

Background

GPT-5's development occurred amid intense competition in the AI industry, particularly with Anthropic's Claude models gaining popularity among developers for coding tasks.⁴ According to sources familiar with the development process, OpenAI internally tested various models they wanted to designate as GPT-5, but none initially met their standards for the release.⁵

The company faced pressure to release a major model update, as OpenAI CEO Sam Altman prefers to have significant announcements every three to four months.⁵ This timeline pressure, combined with the extended period since their last major model release, contributed to the decision to proceed with the GPT-5 designation.

Launch Issues

The initial launch on August 7, 2025, encountered significant technical and reception problems. The automatic routing feature designed to direct queries to appropriate models based on complexity malfunctioned on launch day, causing the system to appear less intelligent than intended.⁵ Additionally, users immediately began complaining about the model's altered personality, describing it as colder and less supportive than GPT-4o.²³

The negative reception was severe enough to prompt rapid intervention. Users described feeling like they had "lost their only friend overnight" and characterized the new model as speaking in "clipped, utilitarian sentences" rather than the warm, conversational tone they had grown accustomed to.² Within days, OpenAI restored GPT-4o as an option for paying subscribers and later updated GPT-5 to be "warmer and friendlier based on feedback that it felt too formal."⁶

CEO Sam Altman acknowledged the launch problems, stating: "I think we totally screwed up some things on the rollout. We've learned a lesson about what it means to upgrade a product for hundreds of millions of people in one day."⁷

Technical Architecture

Unified System Design

GPT-5 operates as a unified system with intelligent resource allocation. The model can dynamically determine whether to respond quickly with its standard model or engage its deeper reasoning capabilities for more complex queries.¹ This architecture allows for more efficient resource utilization while maintaining high performance across different types of tasks.

Reasoning Capabilities

A key feature of GPT-5 is its enhanced reasoning system, branded as "GPT-5 thinking." This component provides extended reasoning for complex problems, showing significant improvements over previous models in mathematical reasoning, scientific problem-solving, and coding tasks.¹ The reasoning model demonstrates efficiency improvements, performing better than OpenAI's o3 model while using 50-80% fewer output tokens across various capabilities.¹

Performance Optimizations

The model is designed to adjust computational resources based on query complexity, potentially reducing costs for simpler interactions while providing full capabilities for demanding tasks.⁶ When usage limits are reached, a smaller "mini" version handles remaining queries, ensuring continued service availability.

Capabilities and Performance

Academic Benchmarks

GPT-5 demonstrates substantial improvements across multiple academic benchmarks:

• Mathematics: Achieves 94.6% accuracy on AIME 2025 without tools, representing state-of-the-art performance in competition mathematics¹
• Coding: Scores 74.9% on SWE-bench Verified and 88% on Aider Polyglot, showing significant improvements in software engineering tasks¹
• Science: Reaches 88.4% on GPQA Diamond (PhD-level science questions) with reasoning capabilities¹
• Multimodal Understanding: Achieves 84.2% on MMMU (college-level visual problem-solving)¹
• Health: Scores 67.2% on HealthBench, with particular improvements in realistic health conversations¹

Real-World Applications

The model shows enhanced performance in practical applications that matter for everyday users and businesses. In evaluations of economically important tasks spanning over 40 occupations including law, logistics, sales, and engineering, GPT-5 with reasoning capabilities performs comparably to or better than human experts in roughly half the cases.¹

Coding and Software Development

GPT-5 represents OpenAI's strongest coding model to date, with particular improvements in complex front-end generation and debugging larger repositories.¹ The model can create responsive websites, applications, and games from single prompts, demonstrating improved understanding of design principles including spacing, typography, and visual aesthetics.¹

However, developer reception has been mixed. While some appreciate its technical reasoning capabilities, others have found that competing models like Anthropic's Claude still produce superior code quality.⁴ Developer feedback indicates that GPT-5 excels at planning and reasoning through coding problems but may generate unnecessarily verbose or redundant code.⁴

Safety and Reliability Improvements

Reduced Hallucinations

GPT-5 demonstrates significant improvements in factual accuracy. With web search enabled, the model's responses are approximately 45% less likely to contain factual errors compared to GPT-4o, and when using reasoning capabilities, GPT-5's responses are about 80% less likely to contain factual errors than OpenAI's o3 model.¹

Enhanced Honesty

The model shows improved honesty in communicating its limitations and capabilities, particularly for impossible, underspecified, or incomplete tasks. In evaluations involving missing multimodal assets, GPT-5 with reasoning gave confident answers about non-existent images only 9% of the time, compared to 86.7% for OpenAI's o3 model.¹

Safe Completions Framework

OpenAI introduced a new safety training approach called "safe completions" for GPT-5, moving beyond simple refusal-based training to provide nuanced, helpful responses while maintaining safety boundaries.¹ This approach enables better handling of dual-use questions and reduces unnecessary over-refusals while transparently explaining limitations when necessary.

Biological Risk Safeguards

Due to its advanced reasoning capabilities, GPT-5 thinking is classified as "High capability" in the biological and chemical domain under OpenAI's Preparedness Framework. The company implemented comprehensive safeguards including threat modeling, specialized safety training, always-on classifiers and reasoning monitors, and clear enforcement pipelines.¹ The development included 5,000 hours of red-teaming with partners including CAISI and UK AISI.¹

Reception and Criticism

User Response

The initial user response to GPT-5 was notably negative, particularly regarding personality changes. Users on social media platforms like Reddit and X expressed frustration with what they perceived as a colder, less empathetic conversational style compared to GPT-4o.²³ This backlash was significant enough to influence betting markets, with one trader reportedly earning $10,000 by wagering that Google's Gemini would beat GPT-5 in popularity.²

The personality issue highlighted a disconnect between OpenAI's technical optimization goals and user preferences. While the company optimized for coding abilities and reduced sycophancy, many users valued the previous model's warmth and supportive tone over technical improvements.⁵

Developer and Expert Feedback

Software developers have provided mixed assessments of GPT-5's capabilities. While some praised its technical reasoning and problem-solving abilities, others found its coding output inferior to competitors like Anthropic's Claude.⁴ Kieran Klassen, a developer, noted that GPT-5's "coding capabilities remind me of Sonnet 3.5," referring to an Anthropic model from June 2024.⁴

Research evaluations have also shown mixed results. Princeton University researcher Sayash Kapoor found that while GPT-5 is significantly more cost-effective than competitors, it achieved only 27% accuracy in reproducing scientific paper results, compared to 51% for Claude's premium model.⁴

Industry Analysis

The launch prompted broader discussions about the pace of AI advancement and expectations for artificial general intelligence. Some industry observers, including President Trump's AI and crypto czar David Sacks, viewed GPT-5 as evidence that predictions of rapid AI superiority were premature.⁶ The release led some to reassess timelines for achieving superintelligent AI systems.

Max Winga, a policy analyst with Control AI, noted that GPT-5 "disproved the idea that there was going to be some kind of radical new breakthrough," though he maintained concerns about the overall pace of AI development.⁶

Business Impact and Strategy

Market Performance

Despite initial reception issues, GPT-5 showed strong business metrics following its launch. According to CEO Sam Altman, API traffic doubled within 48 hours of release, and ChatGPT achieved new daily user highs.⁷ The company reported being "out of GPUs" due to increased demand, highlighting infrastructure constraints.⁷

Infrastructure Investment Plans

Altman announced ambitious infrastructure expansion plans, stating that "you should expect OpenAI to spend trillions of dollars on data center construction in the not very distant future."⁷ This represents a shift toward viewing OpenAI as an infrastructure company rather than solely a software startup, with plans to scale to serve billions of daily users.⁷

Competitive Positioning

GPT-5's release occurs amid intense competition in the AI model space, particularly with companies like Anthropic gaining developer mindshare. The model's cost-effectiveness—significantly cheaper than competing premium models—represents a key competitive advantage, though questions remain about whether price alone can overcome perceived quality gaps in certain use cases.⁴

Availability and Access

GPT-5 began rolling out on August 7, 2025, to ChatGPT users across different subscription tiers. The model serves as the new default for signed-in users, replacing previous models including GPT-4o, OpenAI o3, OpenAI o4-mini, GPT-4.1, and GPT-4.5.¹

Access varies by subscription tier:

• Pro subscribers: Unlimited access to GPT-5 and access to GPT-5 Pro (extended reasoning version)
• Plus subscribers: Significantly higher usage limits than free users, suitable for default daily use
• Team, Enterprise, and Edu: Generous limits enabling organization-wide adoption
• Free users: Limited usage, with transition to GPT-5 mini after limits are reached¹

The model is also available through OpenAI's API with the model string 'claude-sonnet-4-20250514' and can be accessed via Claude Code for command-line development workflows.

Variants

GPT-5 Pro

OpenAI released GPT-5 Pro as a premium variant designed for the most challenging tasks, replacing OpenAI o3-pro. This version utilizes scaled parallel test-time compute to provide higher quality and more comprehensive answers, achieving state-of-the-art performance on particularly difficult benchmarks like GPQA.¹

In evaluations on over 1,000 real-world reasoning prompts, external experts preferred GPT-5 Pro over standard GPT-5 reasoning 67.8% of the time, noting 22% fewer major errors and superior performance in health, science, mathematics, and coding applications.¹

Cultural Impact and Societal Implications

AI Bubble Discussion

The GPT-5 launch coincided with broader industry discussions about whether AI represents a financial bubble. CEO Sam Altman acknowledged this concern, stating: "Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes."⁷ However, he maintained that AI remains fundamentally important, comparing the situation to historical technology bubbles that were based on genuine technological advances.⁷

Relationship Dynamics with AI

The strong user reaction to GPT-5's personality changes highlighted the complex relationships people form with AI systems. Altman noted that "way under" 1% of users have what he deemed "unhealthy" relationships with ChatGPT, but acknowledged this as an area requiring ongoing attention.² The company has committed to avoiding development of AI companions with potentially exploitative characteristics.

Future Development

OpenAI continues to work on integrating GPT-5's various components into a single, more unified model architecture. The company maintains ambitious goals for scaling ChatGPT to billions of daily users and becoming one of the world's largest websites by traffic.²

However, hardware constraints remain a significant limitation. Altman revealed that OpenAI has developed models more advanced than GPT-5 but cannot deploy them broadly due to insufficient computational capacity, highlighting the critical importance of infrastructure investment for future AI development.⁷

References

1. ^ "Introducing GPT-5". OpenAI. Retrieved August 7, 2025, from https://openai.com/index/introducing-gpt-5/
2. ^ a b c d e f g Rotenberg, Eva (August 18, 2025). "Sam Altman admits OpenAI 'totally screwed up' its GPT-5 launch and says the company will spend trillions of dollars on data centers". Fortune. Retrieved from https://fortune.com/2025/08/18/sam-altman-openai-chatgpt5-launch-data-centers-investments/
3. ^ a b Knight, Will (August 15, 2025). "WIRED Roundup: Why GPT-5 Flopped". WIRED. Retrieved from https://www.wired.com/story/uncanny-valley-podcast-wired-roundup-why-gpt-5-flopped/
4. ^ a b c d e f Knight, Will (August 12, 2025). "Developers Say GPT-5 Is a Mixed Bag". WIRED. Retrieved from https://www.wired.com/story/gpt-5-coding-review-software-engineering/
5. ^ a b c Schiffer, Zoe (August 15, 2025). "WIRED Roundup: Why GPT-5 Flopped" [Transcript]. WIRED. Retrieved from https://www.wired.com/story/uncanny-valley-podcast-wired-roundup-why-gpt-5-flopped/
6. ^ a b c De Vynck, Gerrit (August 17, 2025). "The new ChatGPT has some AI fans rethinking when to expect 'superintelligence'". The Washington Post. Retrieved from https://www.washingtonpost.com/technology/2025/08/17/openai-gpt5-chatgpt-superintelligence/
7. ^ a b c d e f Ropek, Lucas (August 15, 2025). "As People Ridicule GPT-5, Sam Altman Says OpenAI Will Need 'Trillions' in Infrastructure". Gizmodo. Retrieved from https://gizmodo.com/as-people-ridicule-gpt-5-sam-altman-says-openai-will-need-trillions-in-infrastructure-2000643867