GPT-5 Statistics 2026: Benchmarks, Adoption and Performance

Last Updated: August 19, 2026

GPT-5 launched August 7, 2025 with 94.6% on AIME 2025 math benchmarks, 74.9% on SWE-bench Verified coding, and 88.4% on GPQA Diamond PhD-level science - with reasoning workloads on OpenAI's platform increasing eightfold within one week of launch per OpenAI's official GPT-5 introduction - while 5 million paid business users adopted GPT-5 at launch with partners including Morgan Stanley, BNY, Figma, Lowe's, and T-Mobile, and by January 2026 only 0.1% of daily ChatGPT users still actively selected the previous GPT-4o model per Feedough's OpenAI statistics tracker. GPT-5 has since evolved into a complete model family - GPT-5.1 through GPT-5.6 - with GPT-5.5 reaching 98.0% on AIME 2025 and 85.0% on ARC-AGI-2 in April 2026, and GPT-5.6 Luna becoming the new default for free users in June 2026.

GPT-5 is not a single model. It is a living product line that has shipped six major versions in under twelve months - the fastest model iteration cadence in AI history. Understanding GPT-5 statistics in August 2026 requires understanding the full family: which model is the current flagship, how each version improved on the last, and how GPT-5 compares to Claude, Gemini, and Grok on the benchmarks that matter for real business decisions.

This guide covers every significant GPT-5 statistic for August 2026 - launch data, benchmark performance across every version, adoption metrics, pricing, competitive comparison, enterprise impact, and EU regulatory compliance - with every data point linked to a named primary source.

🎯 Before you read on - we put together a free 2026 AI Tools Cheat Sheet covering the tools business leaders are actually using right now. Get it instantly when you subscribe to AI Business Weekly.

Get the Free Cheat Sheet

Sources: OpenAI GPT-5 official announcement, MasterOfCode ChatGPT statistics January 2026, Feedough OpenAI statistics 2026, LitSLink GPT-5 guide July 2026, Kingy AI GPT-5.5 benchmarks April 2026

GPT-5 launched on August 7, 2025 - five days after the EU AI Act's general-purpose AI obligations took effect - with OpenAI reporting 5 million paid business users at launch, reasoning workloads increasing eightfold within one week, and by January 2026 only 0.1% of daily ChatGPT users still actively selected the previous GPT-4o model per Feedough's comprehensive OpenAI statistics tracker.

GPT-5 launch statistics:

Sources: Feedough OpenAI statistics 2026, MasterOfCode ChatGPT statistics, Second Talent ChatGPT statistics June 2026

The 8x reasoning workload increase:

Within one week of GPT-5's August 7, 2025 launch, reasoning workloads on OpenAI's platform increased eightfold per Feedough's OpenAI statistics tracker. This is the most significant adoption signal from GPT-5's launch - not just more users, but a fundamental shift in how users engaged with the platform. GPT-5's built-in reasoning capability - automatically routing between fast response and deep thinking modes - changed user behavior immediately. Tasks that users previously would not have submitted to an AI model (complex multi-step reasoning, research synthesis, technical problem solving) became viable with GPT-5, driving the volume surge.

The 0.1% GPT-4o retention finding:

By January 2026, only 0.1% of daily ChatGPT users still actively selected GPT-4o - making it the fastest model obsolescence in AI history. For context: when GPT-4 launched in March 2023, GPT-3.5 remained widely used for months because the capability gap was not sufficient to justify the cost difference. GPT-5's superior performance across every dimension - reasoning, coding, multimodal, hallucination reduction - produced near-complete platform migration in under five months.

The launch partner signal:

BNY Mellon, Morgan Stanley, and Lowe's as day-one enterprise launch partners signals the tier of organizational buyer that OpenAI prioritized for GPT-5 deployment. All three represent high-value, high-volume enterprise relationships where GPT-5's improved reasoning and reduced hallucination rates directly address the verification requirements that enterprise deployments demand. For our complete OpenAI company data, our OpenAI statistics guide covers the full company picture.

GPT-5 achieved 94.6% on AIME 2025 math without tools, 74.9% on SWE-bench Verified coding, 88.4% on GPQA Diamond PhD-level science, and 84.2% on MMMU multimodal benchmarks at launch per OpenAI's official GPT-5 introduction - setting new state-of-the-art results across math, coding, multimodal understanding, and health simultaneously.

GPT-5 original launch benchmark statistics:

Sources: OpenAI official GPT-5 announcement, Runbear GPT-5 benchmarks May 2026, Vellum GPT-5 benchmarks

The AIME 2025 milestone in context:

GPT-5 scoring 94.6% on AIME 2025 without tools - and GPT-5 Pro with Python tools achieving a perfect 100% - is the math benchmark result that most clearly signals a qualitative shift in AI reasoning capability per Nexos AI's benchmark analysis. AIME is the American Invitational Mathematics Examination, a high-difficulty math competition that filters the top 2-3% of competition math participants. Human performance at the 94.6% level represents elite mathematical ability. For GPT-5.2 to subsequently make AIME 2025 essentially "saturated" - meaning the benchmark can no longer meaningfully differentiate between top models - in less than six months of the family's existence indicates the pace of capability improvement.

The SWE-bench Verified coding result:

GPT-5's 74.9% on SWE-bench Verified at launch made it the strongest coding model on release - surpassing o3's 69.1% and GPT-4o's 30.8% by a significant margin per OpenAI's official system card. SWE-bench Verified is the most credible coding benchmark because it uses real GitHub issues from open-source software repositories rather than isolated algorithmic problems. Claude's subsequent performance advantage at SWE-bench (Claude Sonnet 4.6 at 79.6%) reflects Anthropic's specific investment in coding quality - the two models are competitive with each other on coding while diverging more significantly on other tasks.

The hallucination reduction statistics:

GPT-5 producing approximately 45% fewer hallucinations than GPT-4o on web-search-enabled prompts, with thinking enabled reducing hallucinations by approximately 80%, is the enterprise adoption statistic that matters most for organizations using AI for knowledge work per MasterOfCode's January 2026 analysis. Hallucination reduction from 45% to 80% reflects two different mechanisms: the 45% reduction comes from GPT-5's improved factual training, while the 80% reduction with thinking enabled reflects the reasoning layer's ability to verify and self-correct before outputting a response.

For our complete ChatGPT data including the full performance picture, our ChatGPT statistics guide covers every metric.

GPT-5 is not a single model but a rapidly evolving family that shipped six major versions between August 2025 and June 2026 - the fastest model iteration cadence in AI history - with each version improving on specific capability dimensions per LitSLink's July 2026 comprehensive GPT-5 guide and Kingy AI's GPT-5.5 benchmark analysis.

The GPT-5 family timeline:

Sources: LitSLink July 2026, Kingy AI April 2026, Digital Applied April 2026

GPT-5.2 - The first perfect AIME score:

GPT-5.2's perfect AIME score made it the first model to reach 100% on that benchmark through pure model capability per Feedough's tracker. This achievement - occurring less than five months after GPT-5's launch - is what benchmark analysts mean when they describe AIME 2025 as "saturated." Once a model achieves 100% through model capability alone, the benchmark can no longer distinguish between models at the top of the capability range.

GPT-5.5 - The April 2026 capability jump:

GPT-5.5 launched April 23, 2026 with OpenAI framing it as "a new class of intelligence for real work and powering agents" per The Verge's launch coverage cited by Kingy AI. The benchmark numbers that justify that framing: AIME 2025 jumping from GPT-5.4's 92.8% to 98.0%, ARC-AGI-2 reaching 85.0% which is 11.7 points above GPT-5.4 and leads Claude Opus 4.7's 75.8%, and a 1 million token context window. Internal OpenAI adoption of GPT-5.5 reached 85% of the company's own workflows per Kingy AI's analysis.

The pricing architecture across the family:

The GPT-5 family spans a 100x price range from cheapest to most expensive input token per Feedough's analysis - the widest pricing spread in any AI model family. GPT-5 Nano (96% cheaper than standard GPT-5) enables high-volume, cost-sensitive applications that GPT-5's standard pricing would make economically unviable. The pricing architecture reflects OpenAI's strategy: compress costs at the bottom to drive volume adoption, maintain premium pricing at the top to fund capability development.

For the full AI pricing landscape including GPT-5 in context against Claude and Gemini, our AI pricing guide 2026 covers every model and tier.

75% of workers using GPT-5 reported improved speed or quality, saving 40-60 minutes per day on average per MasterOfCode's January 2026 enterprise survey, while reasoning workloads increased 8x in one week of launch and ChatGPT's user base grew from 700 million weekly active users at GPT-5's August 2025 launch to 900 million by February 2026 - a 29% increase attributable significantly to GPT-5's broader capability.

GPT-5 adoption statistics:

Sources: MasterOfCode January 2026, Reuters February 2026, OpenAI enterprise report via Getpanto

The 40-60 minute daily time savings:

75% of workers reporting improved speed or quality with GPT-5, saving 40-60 minutes per day on average, represents the largest worker productivity gain measured from any single AI model update. Across a 250-day working year, 50 minutes per day saved equals approximately 208 hours per worker annually - more than five full working weeks. For an organization of 1,000 workers, that is 208,000 work hours recovered annually from a single model upgrade.

The 9 million paying business users milestone:

Business users growing from 5 million at GPT-5's launch to 9 million+ by mid-2026 in under twelve months reflects the fastest enterprise AI subscription growth in any platform's history per OpenAI's own figures. The 4x growth in under six months documented by Second Talent's June 2026 analysis - across over 7 million enterprise workplace seats - shows GPT-5 driving genuine enterprise deployment rather than just evaluation pilots.

The Custom GPTs and Projects 19x increase:

Custom GPTs and structured workflow Projects increasing 19x year-to-date per OpenAI's enterprise report is the adoption signal that indicates behavioral change rather than casual use. Users building Custom GPTs are integrating GPT-5 into their operational workflows - not experimenting with the model but embedding it in processes that repeat daily. For our complete AI adoption data, our AI adoption statistics guide covers the full enterprise AI adoption picture.

GPT-5's pricing architecture spans a 100x range from GPT-5 Nano (96% cheaper than standard) to GPT-5 Pro at premium pricing - with standard GPT-5 API input at $1.25 per million tokens and output at $10 per million tokens, significantly undercutting Claude Opus 4.1 at $15/$75 per million tokens per LitSLink's July 2026 pricing analysis.

GPT-5 family pricing:

Sources: LitSLink July 2026, Digital Applied April 2026, OpenAI pricing page

The pricing competitive advantage:

GPT-5's standard API pricing at $1.25/$10 per million tokens versus Claude Opus 4.1's $15/$75 is a 12x input price advantage and 7.5x output price advantage per LitSLink's July 2026 analysis. For organizations building AI products at scale, where inference cost multiplies by query volume, this pricing differential is commercially decisive. A product handling one million API calls per day at 2,000 tokens per call costs approximately $2,500 daily on GPT-5 versus $30,000 on Claude Opus 4.1 - a $10 million annual cost difference.

The 100x pricing spread strategy:

The GPT-5 family's 100x range from cheapest to most expensive input token - with Nano at 96% discount and Pro at premium pricing - is the most aggressive model family pricing architecture in AI per Feedough's analysis. OpenAI's strategy: Nano drives adoption in high-volume, cost-sensitive applications that would never use premium models; standard GPT-5 serves enterprise workflows; Pro justifies premium pricing for extended reasoning tasks. The spread serves every price point simultaneously without requiring separate models for different customer segments.

For our complete AI pricing comparison across every major model, our AI pricing guide 2026 covers the full landscape.

GPT-5 leads on math benchmarks and general reasoning, Claude leads on coding quality and long-context tasks, and Gemini 2.5 Pro leads on context window capacity - with the competitive picture in August 2026 showing no single model dominant across all dimensions per multiple independent benchmark analyses.

GPT-5 vs Claude vs Gemini benchmark comparison:

Sources: Second Talent June 2026, Runbear GPT-5 May 2026, Kingy AI April 2026

Where GPT-5 leads:

Math and reasoning benchmarks where GPT-5's unified reasoning system produces consistent advantages. General instruction following where COLLIE's 99.0% accuracy reflects GPT-5's strong performance on diverse task types. Pricing where GPT-5's $1.25 input pricing at 400K context significantly undercuts Claude Opus-tier pricing while delivering comparable capability.

Where Claude leads:

Coding quality where Claude Sonnet 4.6's 79.6% SWE-bench Verified surpasses GPT-5's 74.9%. Long-context reasoning where Claude's 200K context window handles large documents more reliably. Enterprise revenue where Claude surpassed OpenAI's enterprise revenue in mid-2025 per Second Talent's analysis - reflecting Claude's stronger positioning in regulated industries and professional services.

Where Gemini leads:

Context window at 2 million tokens on Gemini Advanced versus GPT-5's 400K via API. Google ecosystem integration through Gmail, Google Drive, and Google Workspace. The Apple Siri integration shipping with iOS 27 that will embed Gemini on 1.4 billion iPhones.

The honest enterprise verdict:

Most serious enterprise teams run both GPT-5 and Claude rather than choosing one per Second Talent's June 2026 analysis. ChatGPT for consumer-facing features with very high weekly usage. Claude for internal engineering workflows, long-context reasoning, and coding. The benchmarks matter for evaluating specific use cases. They do not produce a universal winner for all organizational needs in 2026.

For our complete ChatGPT vs Claude comparison, our ChatGPT vs Claude guide covers every use case executives and developers care about.

Companies deploying ChatGPT Enterprise achieve an average 4.1x ROI within 6 months per Forrester's 2026 Total Economic Impact study, enterprise message volume in ChatGPT Enterprise increased 8x year-over-year, and 92% of Fortune 500 companies are now ChatGPT customers with over 7 million enterprise workplace seats deployed per OpenAI's State of Enterprise AI 2025.

GPT-5 enterprise impact statistics:

Sources: Omnibound ChatGPT user statistics June 2026, Getpanto OpenAI statistics, Second Talent June 2026

The 4.1x Forrester ROI finding:

Forrester's 2026 Total Economic Impact study finding that companies deploying ChatGPT Enterprise achieve 4.1x ROI within 6 months is the enterprise investment justification that CFOs and procurement teams use when evaluating AI licensing decisions. The ROI calculation encompasses time savings across all user workflows, productivity improvements in high-value professional tasks, and cost avoidance from automating work previously done by external vendors.

The $50 million enterprise deal signal:

Top-end enterprise deals above $50 million annually with accounts including Walmart and JPMorgan per Value Add VC reflects the tier of organizational commitment that GPT-5's expanded capabilities unlocked. At sub-GPT-5 capability levels, most enterprises ran limited AI pilots with contained deployments. GPT-5's reasoning capability and hallucination reduction made enterprise-wide deployment viable for organizations where reliability and accuracy are non-negotiable requirements.

The 11.4% referral conversion rate:

ChatGPT referral traffic converting at 11.4% - outperforming every traditional digital marketing channel per ECDB research using Similarweb data - is the commercial signal that explains why enterprises prioritize appearing in AI search results. Traffic arriving from ChatGPT recommendations converts at nearly four times the rate of average organic search traffic. For our complete AI ROI data, our AI ROI statistics guide covers every benchmark.

GPT-5 launched on August 7, 2025 - five days after the EU AI Act's general-purpose AI obligations took effect - without the required training-data summary and copyright policy, with enforcement beginning August 2026 and fines reaching €35 million or 7% of global turnover per LitSLink's July 2026 regulatory analysis.

GPT-5 regulatory statistics:

The compliance timing issue:

GPT-5 shipping five days after the EU AI Act's general-purpose AI obligations took effect - without the required training-data summary and copyright policy - is the regulatory compliance gap that EU-based enterprises must factor into their GPT-5 standardization decisions per LitSLink's July 2026 analysis. Compliance expert Petar Tsankov's assessment that the situation resembles "a speed limit sign without police on the road" reflects the enforcement lag between obligation effective date (August 2025) and active enforcement (August 2026). For organizations in regulated EU sectors, the August 2026 enforcement start date is the planning deadline.

For our complete AI hiring discrimination and regulatory landscape, our AI hiring discrimination guide covers how the EU AI Act applies to employment AI with the same regulatory framework.

ChatGPT Statistics 2026
The complete ChatGPT data - 900 million weekly users, $25 billion ARR, and the full platform picture that GPT-5 powers.

OpenAI Statistics 2026
The complete OpenAI company data - valuation, revenue, funding, and market position behind GPT-5.

OpenAI Company Statistics
The detailed company metrics including enterprise seat growth, geographic distribution, and competitive positioning.

What Is GPT-5?
The complete GPT-5 explainer - what it is, how it works, and how it compares to previous models.

ChatGPT vs Claude
The head-to-head comparison of GPT-5 and Claude on every use case that matters for business decisions.

AI Pricing Guide 2026
GPT-5's $1.25/1M pricing versus Claude and Gemini across every tier and use case.

ChatGPT vs Gemini
How GPT-5 compares to Gemini 3.1 Pro on benchmarks, pricing, and real-world performance.

AI ROI Statistics 2026
The 4.1x Forrester ROI figure in context against AI returns across all industries and platforms.

AI Adoption Statistics 2026
How GPT-5's 92% Fortune 500 adoption fits in the complete enterprise AI adoption picture.

Generative AI Market Statistics 2026
GPT-5 in the complete generative AI market context including competitive market share data.

What are GPT-5's benchmark scores?
GPT-5 launched August 7, 2025 with verified benchmark scores of 94.6% on AIME 2025 math without tools (with GPT-5 Pro achieving 100% with Python tools), 74.9% on SWE-bench Verified coding, 88.4% on GPQA Diamond PhD-level science without tools (89.4% with thinking), 84.2% on MMMU multimodal understanding, 88% on Aider Polyglot coding, 46.2% on HealthBench Hard, 69.6% on Scale MultiChallenge multi-turn reasoning versus GPT-4o's 40.3%, and 99.0% on COLLIE instruction following. GPT-5 reduced hallucinations by approximately 45% compared to GPT-4o on web-search-enabled prompts, with thinking enabled dropping hallucination rates by approximately 80%. Source: OpenAI official GPT-5 announcement, Runbear GPT-5 benchmarks May 2026

When did GPT-5 launch and how was it adopted?
GPT-5 launched on August 7, 2025 - five days after the EU AI Act's general-purpose AI obligations took effect. OpenAI reported 5 million paid business users at launch with day-one enterprise partners including BNY Mellon, Figma, Intercom, Lowe's, Morgan Stanley, and T-Mobile. Within one week of launch, reasoning workloads on OpenAI's platform increased eightfold. By January 2026, only 0.1% of daily ChatGPT users still actively selected the previous GPT-4o model. OpenAI retired GPT-4o and five other legacy models on February 13, 2026. ChatGPT weekly active users grew from approximately 700 million at GPT-5's launch to 900 million by February 2026 - a 29% increase over six months significantly attributable to GPT-5's broader capability attracting new users and increasing existing user engagement. Source: Feedough OpenAI statistics, MasterOfCode January 2026

What is the GPT-5 family and what versions exist?
GPT-5 is a living product family that shipped six major versions between August 2025 and June 2026. GPT-5 launched August 7, 2025 with the original benchmark results. GPT-5.1 (November 13, 2025) improved writing and data science. GPT-5.2 (December 11, 2025) was the first model to achieve a perfect 100% AIME score through pure model capability. GPT-5.3 and GPT-5.4 shipped in February and March 2026 with iterative improvements. GPT-5.5 (April 23, 2026) reached 98.0% on AIME 2025 and 85.0% on ARC-AGI-2 with a 1 million token context window. GPT-5.6 Luna became the new default for Free and Go users in June 2026. The active lineup as of August 2026 includes GPT-5.5, GPT-5.4, GPT-5.3 Instant, and GPT-5.4 Mini. Source: LitSLink July 2026, Kingy AI April 2026

How does GPT-5 pricing compare to Claude and Gemini?
GPT-5 standard API pricing is $1.25 per million input tokens and $10 per million output tokens. GPT-5 Mini costs approximately 80% less than standard GPT-5 and GPT-5 Nano costs approximately 96% less. GPT-5.5 Pro is at premium pricing above standard GPT-5. By comparison, Claude Opus 4.1 is priced at $15 per million input tokens and $75 per million output tokens - making GPT-5 standard approximately 12x cheaper on input and 7.5x cheaper on output. Gemini 2.5 Pro matches GPT-5 at $1.25 per million input tokens for the 200K context tier. The GPT-5 family spans a 100x price range from cheapest to most expensive per Feedough's analysis - the widest pricing spread of any AI model family, designed to serve every use case from high-volume commodity tasks (Nano) to extended reasoning workloads (Pro). Source: LitSLink July 2026, Feedough

How does GPT-5 compare to Claude on coding?
GPT-5 achieved 74.9% on SWE-bench Verified at launch - surpassing OpenAI's own o3 model at 69.1% and GPT-4o at 30.8%. However Claude Sonnet 4.6 achieves 79.6% on SWE-bench Verified, giving Claude a measurable coding quality advantage over GPT-5 standard. For coding-heavy workloads, this benchmark differential is the primary reason many enterprise engineering teams use Claude for internal development workflows while using GPT-5 for consumer-facing and breadth-of-task applications. Claude also leads on long-context code review with its 200K context window versus GPT-5's practical context limit. Most serious enterprise teams run both: ChatGPT for consumer-facing features and volume, Claude for internal engineering workflows requiring coding depth and long-context reasoning per Second Talent's June 2026 analysis. Source: Second Talent June 2026, OpenAI official

What is GPT-5's enterprise ROI?
Companies deploying ChatGPT Enterprise achieve an average 4.1x return on investment within 6 months per Forrester's 2026 Total Economic Impact study. Enterprise message volume in ChatGPT Enterprise increased 8x year-over-year, average worker message volume increased 30%, and structured workflow usage through Custom GPTs and Projects increased 19x year-to-date per OpenAI's enterprise report. 92% of Fortune 500 companies are ChatGPT customers with over 7 million enterprise workplace seats active - a 9x increase year-over-year. Paying business users grew from 5 million at GPT-5's launch to 9 million+ by mid-2026. Top enterprise contracts reach $50 million annually with accounts including Walmart and JPMorgan per Value Add VC. 75% of workers using GPT-5 reported improved speed or quality, saving 40-60 minutes per day on average. Source: Omnibound June 2026

What is GPT-5.5 and how does it compare to GPT-5?
GPT-5.5 launched April 23, 2026 as OpenAI's smartest and most intuitive model, described by OpenAI as "a new class of intelligence for real work and powering agents." Key benchmark improvements over GPT-5.4: AIME 2025 improved from 92.8% to 98.0% (a 5.2 point gain), ARC-AGI-2 reached 85.0% which is 11.7 points above GPT-5.4 and leads Claude Opus 4.7's 75.8%, and the context window expanded to 1 million tokens. Internal OpenAI adoption of GPT-5.5 reached 85% of the company's own workflows. GPT-5.5 is specifically optimized for agentic workflows - completing multi-step tasks that span multiple tools and sessions - rather than single-turn query answering. For organizations building AI agents or needing the highest available reasoning quality, GPT-5.5 is the current benchmark leader as of August 2026. Source: Kingy AI April 2026

The GPT-5 statistics of August 2026 document the fastest model capability evolution and adoption curve in AI history - simultaneously.

Six model versions in twelve months. Perfect AIME scores. 8x reasoning workload increase in one week. 92% Fortune 500 adoption. 4.1x enterprise ROI in six months. 0.1% of users remaining on the previous model by January 2026. These are the statistics of a technology that did not gradually improve - it leaped.

The competitive picture is the most commercially important context. GPT-5 leads on math benchmarks and pricing. Claude leads on coding quality and long-context tasks. Gemini leads on context window capacity and Google ecosystem integration. No model is universally superior across all dimensions, which explains why most serious enterprise teams run multiple models rather than committing exclusively to one.

The GPT-5.5 trajectory matters for planning. ARC-AGI-2 at 85.0% versus Claude Opus 4.7's 75.8% represents the clearest benchmark leadership GPT-5 has demonstrated over Claude since launch. Internal OpenAI adoption of GPT-5.5 at 85% of their own workflows is the strongest signal that GPT-5.5 delivers material improvement over earlier versions in production conditions. Whether GPT-5.6 Luna, now the default for free users, brings those capabilities to the broadest possible audience will define ChatGPT's user growth trajectory through the end of 2026.

The EU regulatory gap - GPT-5 launching without required training data documentation five days after EU AI Act obligations took effect, with enforcement beginning August 2026 - is the compliance planning item for every European enterprise standardizing on GPT-5. The enforcement clock is running.