{
  "slug": "ai-coding-models-statistics-2026",
  "title": "AI Coding Models Statistics 2026",
  "compiled": "2026-04-25",
  "verified": "2026-04-25",
  "compiler": "Vincent Forat / Preuve AI",
  "verification_method": "4 parallel web-explorer subagents, each cross-checked URLs against primary publishers. Sources marked verified=true had their stat located in the cited URL. Sources that could not be verified to a primary publisher were REMOVED from the article body.",
  "note": "Page renders ZERO outbound <a> tags by design (Qwoted SEO playbook: hoard PageRank). This file is the audit trail of where each number came from.",
  "removed_during_verification": [
    {
      "claim": "Claude Opus 4.6 scores 83.7% on SWE-bench Verified",
      "reason": "Outdated. Anthropic released Opus 4.7 on April 16, 2026 with 87.6% SWE-bench Verified. Article body and FAQ updated.",
      "replaced_with": "Claude Opus 4.7 at 87.6% on SWE-bench Verified"
    },
    {
      "claim": "GPT-5.4 scores 80% on SWE-bench Verified; GPT-5 leads Aider Polyglot at 88.0% pass-rate-2",
      "reason": "Could not verify GPT-5.4 specific score. Aider leaderboard does not show GPT-5 at 88%; OpenAI's top reasoning model on Aider is in the 80s range.",
      "replaced_with": "OpenAI's reasoning models cluster in the high 70s to low 80s in publicly reported figures"
    },
    {
      "claim": "Gemini 3 Pro Preview leads LiveCodeBench at 91.7%",
      "reason": "Specific score not verifiable on the LiveCodeBench leaderboard at verification time.",
      "replaced_with": "Removed; replaced with general statement that Gemini Pro and DeepSeek trade the top spot"
    },
    {
      "claim": "Gemini 2.5 Pro scores 83.1% on Aider Polyglot",
      "reason": "Verified score is closer to 76.5%; cited number could not be reproduced.",
      "replaced_with": "Removed"
    },
    {
      "claim": "Grok 4 scores 79.6% on Aider Polyglot",
      "reason": "Could not verify Grok-4 model release or its leaderboard entry. Latest publicly confirmed xAI model at verification time was Grok-2/3.",
      "replaced_with": "Removed"
    },
    {
      "claim": "DeepSeek V3.2 scores 74.2% on Aider Polyglot; 0.833 on LiveCodeBench Thinking mode",
      "reason": "DeepSeek V3.2 model name could not be verified. DeepSeek V3 (December 2024) is the verified release.",
      "replaced_with": "DeepSeek V3 referenced as strongest open-weights result"
    },
    {
      "claim": "GPT-5.5 pricing: $5 input / $18 output per 1M tokens",
      "reason": "Verified OpenAI pricing for GPT-5.5 is $5 / $30, not $5 / $18. Pricing table updated.",
      "replaced_with": "GPT-5.5: $5 input / $30 output per 1M tokens"
    },
    {
      "claim": "32% more PRs merged per developer per week (GitHub-Accenture study)",
      "reason": "Verified GitHub-Accenture study reports 8.69% PR increase and 15% merge rate increase, NOT 32%.",
      "replaced_with": "Removed; productivity section now leads on MIT 26.08% finding"
    },
    {
      "claim": "30-40% productivity gains for developers using gen AI (McKinsey citing IBM Software internal data)",
      "reason": "Could not be traced to a primary McKinsey or IBM publication. Likely secondary aggregation.",
      "replaced_with": "Removed; replaced with verified McKinsey 20-45% range from Economic Potential of Generative AI report"
    },
    {
      "claim": "90% of developers report time savings, median ~20% (arxiv 2501.13282)",
      "reason": "The actual Zoominfo paper (arxiv 2501.13282) reports 33% suggestion acceptance, 20% line acceptance, 72% developer satisfaction, NOT 90% time savings.",
      "replaced_with": "Updated to use the paper's actual findings"
    },
    {
      "claim": "73% of engineering teams use AI coding tools daily, up from 41% (Developer Ecosystem Research Group, 2026)",
      "reason": "'Developer Ecosystem Research Group' could not be verified as a real publishing organization. Stat sourced only via secondary aggregator (Derivinate News).",
      "replaced_with": "Removed; replaced with verified Stack Overflow 2024 51% daily-use figure"
    },
    {
      "claim": "By 2028, autonomous AI agents will handle 60-70% of routine development tasks (Forrester Predictions 2026)",
      "reason": "Could not locate the specific Forrester Predictions 2026 post containing this number.",
      "replaced_with": "Removed; future projections section now relies on verified Gartner forecast"
    },
    {
      "claim": "42% of AI teams employ 2+ LLMs in production (State of AI 2025 Survey)",
      "reason": "Stat could not be verified in the public State of AI 2025 report.",
      "replaced_with": "Removed; multi-LLM section relies on verified Datadog and DeepSense data"
    },
    {
      "claim": "20% of AI-suggested package names do not exist; 43% repeat consistently (Socket Security)",
      "reason": "Verified primary source (Spracklen et al., USENIX Security 2025) reports 19.7% hallucination rate, 43% similar to real names, 38% repeat consistently. Updated for accuracy.",
      "replaced_with": "Updated to use exact USENIX Security 2025 figures"
    }
  ],
  "verified_sources": [
    {
      "id": "stackoverflow-survey-2024",
      "publisher": "Stack Overflow",
      "title": "AI - 2024 Stack Overflow Developer Survey",
      "year": 2024,
      "url": "https://survey.stackoverflow.co/2024/ai",
      "verified": true,
      "stats_used": [
        "76% of developers use or plan to use AI tools (up from 70% in 2023)",
        "62% currently use AI tools (up from 44%)",
        "84% of pros have AI in workflow, 51% use daily",
        "85% of pros use AI for writing code"
      ]
    },
    {
      "id": "stackoverflow-distrust-2024",
      "publisher": "Stack Overflow Blog",
      "title": "Where developers feel AI coding tools are working - and where they're missing the mark",
      "year": 2024,
      "url": "https://stackoverflow.blog/2024/09/23/where-developers-feel-ai-coding-tools-are-working-and-where-they-re-missing-the-mark",
      "verified": true,
      "stats_used": [
        "66% cite distrust as top concern",
        "63% cite lack of codebase context",
        "42% trust AI accuracy, 31% explicitly distrust"
      ]
    },
    {
      "id": "stackoverflow-trust-2026",
      "publisher": "Stack Overflow Blog",
      "title": "Why AI hasn't replaced human expertise (April 2026)",
      "year": 2026,
      "url": "https://stackoverflow.blog/2026/04/15/why-ai-hasn-t-replaced-human-expertise/",
      "verified": true,
      "stats_used": [
        "29% of developers trust AI tools in 2026 (down 11pp from 40% in 2024); 46% actively distrust"
      ]
    },
    {
      "id": "github-octoverse-2024",
      "publisher": "GitHub",
      "title": "Octoverse 2024",
      "year": 2024,
      "url": "https://github.blog/news-insights/octoverse/octoverse-2024/",
      "verified": true,
      "stats_used": [
        "98% YoY growth in generative AI projects",
        "59% surge in AI-related contributions",
        "70,000+ new public AI projects"
      ]
    },
    {
      "id": "github-peng-2022",
      "publisher": "GitHub Blog",
      "title": "Quantifying GitHub Copilot's impact on developer productivity and happiness (Peng et al.)",
      "year": 2022,
      "url": "https://github.blog/news-insights/research/research-quantifying-github-copilots-impact-on-developer-productivity-and-happiness/",
      "verified": true,
      "stats_used": [
        "55% faster task completion in controlled lab experiment",
        "78% completion rate vs 70% control"
      ]
    },
    {
      "id": "github-accenture-2024",
      "publisher": "GitHub Blog",
      "title": "Research: Quantifying GitHub Copilot's impact in the enterprise with Accenture",
      "year": 2024,
      "url": "https://github.blog/2024-05-13-research-quantifying-github-copilots-impact-in-the-enterprise-with-accenture/",
      "verified": true,
      "stats_used": [
        "85% of developers more confident in code quality with Copilot",
        "Real PR-merge effect: 8.69% PR increase, 15% merge-rate increase (NOT 32% as originally written)"
      ],
      "correction_note": "Article previously claimed 32% more PRs merged - this number is not in the GitHub-Accenture study. Removed from body."
    },
    {
      "id": "github-code-quality-2024",
      "publisher": "GitHub Blog",
      "title": "Does GitHub Copilot improve code quality? Here's what the data says",
      "year": 2024,
      "url": "https://github.blog/news-insights/research/does-github-copilot-improve-code-quality-heres-what-the-data-says/",
      "verified": true,
      "stats_used": [
        "13.6% more lines of code per error with Copilot (18.2 vs 16.0 lines per error)"
      ]
    },
    {
      "id": "github-microsoft-fy2026",
      "publisher": "Microsoft / GitHub (FY2026 earnings)",
      "title": "GitHub Copilot 20M users / 4.7M paid",
      "year": 2025,
      "url": "https://dev.ua/en/news/github-copilot-1753946221",
      "verified": true,
      "url_note": "Secondary news aggregator - the original Microsoft earnings disclosure is the primary source. Verify Microsoft IR page for exact filing.",
      "stats_used": [
        "20M+ all-time GitHub Copilot users by July 2025",
        "4.7M paid Copilot subscribers, up 75% YoY",
        "90% of Fortune 100 deployed Copilot"
      ]
    },
    {
      "id": "jetbrains-devecosystem-2024",
      "publisher": "JetBrains",
      "title": "State of Developer Ecosystem 2024",
      "year": 2024,
      "url": "https://www.jetbrains.com/lp/devecosystem-2024/",
      "verified": true,
      "stats_used": [
        "69% tried ChatGPT for coding, 49% regular use",
        "40% tried GitHub Copilot, 26% regular use",
        "80% of companies allow third-party AI tools or have no formal restrictions"
      ]
    },
    {
      "id": "anthropic-opus-47",
      "publisher": "Anthropic",
      "title": "Claude Opus 4.7 (April 16, 2026 release)",
      "year": 2026,
      "url": "https://www.anthropic.com/pricing",
      "verified": true,
      "stats_used": [
        "Claude Opus 4.7: 87.6% on SWE-bench Verified",
        "Claude Opus 4.7 pricing: $5 input / $25 output per 1M tokens"
      ]
    },
    {
      "id": "swebench-leaderboard",
      "publisher": "Princeton NLP / SWE-bench team",
      "title": "SWE-bench Verified leaderboard",
      "year": 2026,
      "url": "https://www.swebench.com/",
      "verified": true,
      "stats_used": [
        "Cross-checked SWE-bench Verified scores"
      ]
    },
    {
      "id": "google-gemini-31",
      "publisher": "Google DeepMind",
      "title": "Gemini 3.1 Pro model card",
      "year": 2026,
      "url": "https://deepmind.google/models/gemini/pro/",
      "verified": true,
      "stats_used": [
        "Gemini 3.1 Pro: 80.6% on SWE-bench Verified"
      ]
    },
    {
      "id": "openai-pricing",
      "publisher": "OpenAI",
      "title": "API Pricing",
      "year": 2026,
      "url": "https://platform.openai.com/docs/pricing",
      "verified": true,
      "stats_used": [
        "GPT-5.5: $5 input / $30 output per 1M tokens"
      ],
      "correction_note": "Article body originally listed $5/$18; corrected to verified $5/$30."
    },
    {
      "id": "aider-leaderboard",
      "publisher": "Aider",
      "title": "Aider LLM Leaderboards",
      "year": 2025,
      "url": "https://aider.chat/docs/leaderboards/",
      "verified": true,
      "stats_used": [
        "Aider Polyglot benchmark methodology: 225 Exercism exercises across 6 languages"
      ]
    },
    {
      "id": "livecodebench",
      "publisher": "LiveCodeBench team",
      "title": "LiveCodeBench leaderboard",
      "year": 2026,
      "url": "https://livecodebench.github.io/",
      "verified": true,
      "stats_used": [
        "Competitive-programming benchmark used to confirm leaderboard cluster"
      ]
    },
    {
      "id": "mit-ssrn-4945566",
      "publisher": "MIT Economics / SSRN",
      "title": "The Effects of Generative AI on High-Skilled Work: Evidence from Three Field Experiments with Software Developers",
      "authors": "Zheyuan Cui, Mert Demirer, Sonia Jaffe, Leon Musolff, Sida Peng, Tobias Salz",
      "year": 2025,
      "url": "https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4945566",
      "verified": true,
      "stats_used": [
        "26.08% increase in completed tasks (SE 10.3%) across n=4,867 developers at Microsoft, Accenture, anonymous Fortune 100 firm"
      ]
    },
    {
      "id": "google-research-arxiv-2410-12944",
      "publisher": "Google Research / arXiv",
      "title": "How much does AI impact development speed? An enterprise-based randomized controlled trial",
      "authors": "Paradis, Grey, Madison, Nam, Macvean, Meimand, Zhang, Ferrari-Church, Chandra",
      "year": 2024,
      "url": "https://arxiv.org/abs/2410.12944",
      "verified": true,
      "stats_used": [
        "21% faster task completion in controlled trial of 96 Google software engineers"
      ]
    },
    {
      "id": "zoominfo-arxiv-2501-13282",
      "publisher": "Zoominfo / arXiv",
      "title": "Experience with GitHub Copilot for Developer Productivity at Zoominfo",
      "authors": "Bakal, Dasdan, Katz, Kaufman, Levin",
      "year": 2025,
      "url": "https://arxiv.org/abs/2501.13282",
      "verified": true,
      "stats_used": [
        "33% suggestion acceptance rate",
        "20% line acceptance rate",
        "72% developer satisfaction"
      ],
      "correction_note": "Article previously claimed '90% of developers report time savings, median 20%' which is not in this paper. Replaced with paper's actual findings."
    },
    {
      "id": "mckinsey-economic-potential-2023",
      "publisher": "McKinsey",
      "title": "The economic potential of generative AI: The next productivity frontier",
      "authors": "Chui, Hazan, Roberts, Singla, Smaje, Sukharevsky, Yee, Zemmel",
      "year": 2023,
      "url": "https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/the-economic-potential-of-generative-ai-the-next-productivity-frontier",
      "verified": true,
      "stats_used": [
        "Software engineering = 20-45% of generative AI's annual economic value"
      ]
    },
    {
      "id": "mckinsey-software-disruption-2024",
      "publisher": "McKinsey",
      "title": "Navigating the generative AI disruption in software",
      "year": 2024,
      "url": "https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/navigating-the-generative-ai-disruption-in-software",
      "verified": true,
      "stats_used": [
        "Generative AI software spending: $175B-$250B by 2027 (2-6 percentage points of additional sector growth)"
      ]
    },
    {
      "id": "ucsd-copilot-2024",
      "publisher": "UC San Diego IT Services",
      "title": "How GitHub Copilot Boosted Developer Productivity",
      "year": 2024,
      "url": "https://blink.ucsd.edu/technology/about/news/posts/2024-08-01-github-copilot.html",
      "verified": true,
      "stats_used": [
        "48% less time on high-complexity tasks",
        "63% less time on low-complexity tasks (8-week internal study)"
      ]
    },
    {
      "id": "veracode-genai-2025",
      "publisher": "Veracode",
      "title": "2025 GenAI Code Security Report",
      "year": 2025,
      "url": "https://www.veracode.com/blog/genai-code-security-report/",
      "verified": true,
      "stats_used": [
        "45% of AI-generated code fails OWASP Top 10",
        "Java 72% failure, Python 38%, JavaScript 43%, C# 45%",
        "86% fail to prevent CWE-80 (XSS)"
      ]
    },
    {
      "id": "veracode-october-2025",
      "publisher": "Veracode",
      "title": "October 2025 Update: GenAI Code Security Report",
      "year": 2025,
      "url": "https://www.veracode.com/resources/analyst-reports/2025-genai-code-security-report/",
      "verified": true,
      "stats_used": [
        "Pass rate flat at ~55% across 2023-2025 despite model upgrades"
      ]
    },
    {
      "id": "spracklen-usenix-2025",
      "publisher": "USENIX Security 2025 / Spracklen et al.",
      "title": "Slopsquatting: AI Hallucinated Package Names in Code Generation",
      "year": 2025,
      "url": "https://nesbitt.io/2025/12/10/slopsquatting-meets-dependency-confusion",
      "url_note": "Secondary writeup; the primary USENIX 2025 paper PDF should be cited when pitching to journalists.",
      "verified": true,
      "stats_used": [
        "19.7% of AI-suggested package names hallucinated (576,000 samples across 16 LLMs)",
        "205,474 unique fake package names",
        "43% similar to real package names",
        "38% repeat consistently across runs"
      ]
    },
    {
      "id": "grand-view-research-2024",
      "publisher": "Grand View Research",
      "title": "AI Code Tools Market Report",
      "year": 2024,
      "url": "https://www.grandviewresearch.com/industry-analysis/ai-code-tools-market-report",
      "verified": true,
      "stats_used": [
        "Market $4.86B (2023) - $26.03B (2030) at 27.1% CAGR"
      ]
    },
    {
      "id": "mordor-intelligence-2025",
      "publisher": "Mordor Intelligence",
      "title": "AI Code Tools Market - Size, Share & 2030 Trends Report",
      "year": 2025,
      "url": "https://www.mordorintelligence.com/industry-reports/artificial-intelligence-code-tools-market",
      "verified": true,
      "stats_used": [
        "Market $7.37B (2025) - $23.97B (2030) at 26.6% CAGR"
      ]
    },
    {
      "id": "sns-insider-2025",
      "publisher": "SNS Insider (via Globe Newswire)",
      "title": "AI Code Tools Market to Hit USD 37.34 Billion by 2032",
      "year": 2025,
      "url": "https://www.globenewswire.com/news-release/2025/09/26/3157060/0/en/ai-code-tools-market-to-hit-usd-37-34-billion-by-2032-driven-by-rising-demand-for-automation-globally-sns-insider.html",
      "verified": true,
      "stats_used": [
        "Market $6.04B (2024) - $37.34B (2032) at 25.62% CAGR (2025-2032)"
      ]
    },
    {
      "id": "datadog-state-ai-engineering-2026",
      "publisher": "Datadog",
      "title": "2026 State of AI Engineering",
      "year": 2026,
      "url": "https://www.datadoghq.com/state-of-ai-engineering/",
      "verified": true,
      "stats_used": [
        "69% of companies run 3+ AI models in production (up from ~40% in 2024)",
        "OpenAI used by 63% of companies",
        "77% use multiple model providers in production"
      ]
    },
    {
      "id": "langchain-state-agent-engineering-2026",
      "publisher": "LangChain",
      "title": "State of Agent Engineering 2026",
      "year": 2026,
      "url": "https://www.langchain.com/state-of-agent-engineering",
      "verified": true,
      "stats_used": [
        "Multi-model architectures are the production norm in 2026"
      ],
      "verification_note": "The 75%+ multi-model figure is consistent with the report's themes but was not stated as a verbatim quote in accessible sections. Use the directional finding, not a hard percentage."
    },
    {
      "id": "deepsense-cto-survey-2025",
      "publisher": "DeepSense.ai",
      "title": "What It Really Takes to Scale LLMs in 2025-26",
      "year": 2025,
      "url": "https://deepsense.ai/resource/what-it-really-takes-to-scale-llms-in-2025-26/",
      "verified": true,
      "stats_used": [
        "65% of enterprise teams prioritize multi-model agentic systems"
      ]
    },
    {
      "id": "sozo-pulse-cto-2024",
      "publisher": "Sōzō (SoftBank Vision Fund)",
      "title": "Pulse CTO Survey 2024 Q3",
      "year": 2024,
      "url": "https://visionfund.com/sozo-pulse/cto-survey-2024-q3",
      "verified": true,
      "stats_used": [
        "63% of CTOs increased gen AI budgets in 2024",
        "24% doubled gen AI budgets or more"
      ]
    },
    {
      "id": "gartner-2024-press",
      "publisher": "Gartner",
      "title": "Gartner Says 75 Percent of Enterprise Software Engineers Will Use AI Code Assistants by 2028 (April 11, 2024 press release)",
      "year": 2024,
      "url": "https://www.gartner.com/en/newsroom/press-releases/2024-04-11-gartner-says-75-percent-of-enterprise-software-engineers-will-use-ai-code-assistants-by-2028",
      "verified": true,
      "stats_used": [
        "75% of enterprise software engineers will use AI code assistants by 2028 (from <10% early 2023)"
      ]
    },
    {
      "id": "stanford-ai-index-2026",
      "publisher": "Stanford HAI",
      "title": "AI Index 2026",
      "year": 2026,
      "url": "https://hai.stanford.edu/ai-index/2026-ai-index-report",
      "verified": true,
      "stats_used": [
        "Security is #1 scaling barrier for agentic AI, cited by 62% of organizations"
      ]
    }
  ]
}