AI Benchmarks: chart data from AI Adoption (https://aiadoption.org/ai-benchmarks/)

Data from the AI Index Report 2026 (Stanford HAI) and the partners named below. Licence: CC BY-ND 4.0, as for the AI Index.
Each CSV is the data behind one chart. Cite the chart page and its original source.

ai-benchmarks-vs-human-baseline.csv  The best SWE-bench Verified score reached 98.7% of the human baseline in 2025, up from 64.3%
    Source: AI Index, 2026
    Chart: https://aiadoption.org/data/ai-benchmarks-vs-human-baseline/
text-arena-leaderboard.csv  Claude Opus 4.6 tops the Text Arena at 1,503, and the top 15 models span 46 points
    Source: Arena, 2026
    Chart: https://aiadoption.org/data/text-arena-leaderboard/
mmlu-pro-leaderboard.csv  Gemini-3.1-Pro leads MMLU-Pro at 91.2%, and the top 15 models are within 4.2 points
    Source: MMLU-Pro Leaderboard, 2026
    Chart: https://aiadoption.org/data/mmlu-pro-leaderboard/
gpqa-diamond-top-accuracy.csv  The best GPQA Diamond score rose from 36% in 2023 to 93% in 2025, above the 81.2% expert baseline
    Source: Epoch AI, 2026
    Chart: https://aiadoption.org/data/gpqa-diamond-top-accuracy/
humanitys-last-exam-top-accuracy.csv  The best score on Humanity's Last Exam rose from 8% in 2024 to 38.3% in 2025
    Source: Center for AI Safety et al., 2026
    Chart: https://aiadoption.org/data/humanitys-last-exam-top-accuracy/
arc-agi-2-leaderboard.csv  Gemini 3 Deep Think scored 84.6% on ARC-AGI-2, 7.5 points ahead of the next system
    Source: ARC-AGI-2 Leaderboard, 2026
    Chart: https://aiadoption.org/data/arc-agi-2-leaderboard/
planbench-standard-planning.csv  GPT-5 solved 205 PlanBench tasks across eight domains, level with the classical planner LAMA's 204
    Source: Corrêa et al., 2025
    Chart: https://aiadoption.org/data/planbench-standard-planning/
planbench-obfuscated-planning.csv  Disguising PlanBench tasks cut DeepSeek R1's solved total from 157 to 93; LAMA still solved 204
    Source: Corrêa et al., 2025
    Chart: https://aiadoption.org/data/planbench-obfuscated-planning/
frontiermath-tier-4-accuracy.csv  The best model solved 31.3% of FrontierMath's hardest problems in 2025, up from 0% in 2024
    Source: Epoch AI, 2026
    Chart: https://aiadoption.org/data/frontiermath-tier-4-accuracy/
matharena-top-accuracy.csv  GPT-5.2 (high) scored 97% on MathArena in December 2025, up from 86.4% for November's best model
    Source: MathArena Leaderboard, 2026
    Chart: https://aiadoption.org/data/matharena-top-accuracy/
imo-proofbench-proof-scores.csv  Aletheia scored 91.9% on IMO-ProofBench olympiad proofs in early 2026, against 7.1% for GPT-5.1
    Source: IMO-ProofBench Leaderboard, 2026
    Chart: https://aiadoption.org/data/imo-proofbench-proof-scores/
swe-bench-verified-top-models.csv  Claude 4.5 Opus led SWE-bench Verified at 76.8% in February 2026; the top 10 sat within 6 points
    Source: SWE-bench Leaderboard, 2026
    Chart: https://aiadoption.org/data/swe-bench-verified-top-models/
terminal-bench-top-score.csv  The best Terminal-Bench 2.0 score rose from 18.4% to 77.3% between February 2025 and February 2026
    Source: Terminal-Bench 2.0 Leaderboard, 2026
    Chart: https://aiadoption.org/data/terminal-bench-top-score/
vibe-code-bench-app-building.csv  Claude Opus 4.6 led Vibe Code Bench at 57.6% in early 2026, 26 points above the lowest of eight models
    Source: Vals.ai, 2026
    Chart: https://aiadoption.org/data/vibe-code-bench-app-building/
berkeley-function-calling-leaderboard.csv  Claude-Opus-4.5 leads the Berkeley Function Calling Leaderboard at 77.5%, 4.2 points clear of the next
    Source: Berkeley Function Calling Leaderboard, 2026
    Chart: https://aiadoption.org/data/berkeley-function-calling-leaderboard/
gaia-ai-assistant-benchmark.csv  The best model scored 74.5% on GAIA in September 2025, 17.5 points short of the 92% human baseline
    Source: GAIA Leaderboard, 2026
    Chart: https://aiadoption.org/data/gaia-ai-assistant-benchmark/
osworld-computer-use-agents.csv  The best AI agent completed 66.3% of OSWorld computer tasks in 2025, against 72.4% for people
    Source: Epoch AI, 2026
    Chart: https://aiadoption.org/data/osworld-computer-use-agents/
webarena-web-agents.csv  Web agents completed 74.3% of WebArena tasks in early 2026, within 4 points of people's 78.2%
    Source: WebArena Leaderboard, 2026
    Chart: https://aiadoption.org/data/webarena-web-agents/
mle-bench-ml-engineering-agents.csv  The best agent succeeded on 64.4% of MLE-bench's Kaggle tasks in early 2026, up from 17.1% in 2024
    Source: MLE-bench Leaderboard, 2026
    Chart: https://aiadoption.org/data/mle-bench-ml-engineering-agents/
cybench-cybersecurity-agents.csv  AI agents solved 93% of Cybench security challenges unguided in early 2026, up from 17.5% in 2024
    Source: Cybench Leaderboard, 2026
    Chart: https://aiadoption.org/data/cybench-cybersecurity-agents/
tau-bench-tool-use-agents.csv  Claude Opus 4.5 led τ-bench's tool-use tasks at 70.2% in early 2026; the top seven span 7.3 points
    Source: τ-bench Leaderboard, 2026
    Chart: https://aiadoption.org/data/tau-bench-tool-use-agents/
helm-arabic-llm-scores.csv  Arabic.AI's LLM-X topped HELM Arabic with a mean score of 0.86, ahead of Gemini 2.5 Flash at 0.82
    Source: HELM, 2026
    Chart: https://aiadoption.org/data/helm-arabic-llm-scores/
indic-llm-arena-scores.csv  GPT-5.2 leads the Indic LLM Arena with 1,314, ahead of GPT-5.1 (1,298) and Gemini 3 Flash (1,288)
    Source: Indic LLM Arena Leaderboard, 2026
    Chart: https://aiadoption.org/data/indic-llm-arena-scores/
slovenian-dialect-llm-accuracy.csv  GPT-5 scored 99.8% on standard Slovenian but 88.6% on the Cerkno dialect; 8 of 12 models fell below 60%
    Source: Slovene DIALECT-COPA benchmark leaderboard, 2026
    Chart: https://aiadoption.org/data/slovenian-dialect-llm-accuracy/
mteb-text-embedding-top-score.csv  The top MTEB English embedding score rose to 76.0 in 2025, from 58.3 in 2019
    Source: MTEB Leaderboard, 2026
    Chart: https://aiadoption.org/data/mteb-text-embedding-top-score/
mmmu-leaderboard.csv  Gemini 3.1 Pro Preview scored 88.2% on MMMU, 0.4 points short of the best human expert
    Source: Vals.ai, 2026
    Chart: https://aiadoption.org/data/mmmu-leaderboard/
vision-arena-leaderboard.csv  Google's gemini-3-pro leads the Vision Arena at 1,288; Gemini models hold four of the top six places
    Source: Arena, 2026
    Chart: https://aiadoption.org/data/vision-arena-leaderboard/
clockqa-analogue-clock-reading.csv  The best of seven AI models read analogue clocks exactly right 22.6% of the time on ClockQA
    Source: Saxena et al., 2025
    Chart: https://aiadoption.org/data/clockqa-analogue-clock-reading/
calendarqa-date-reasoning.csv  GPT-o1 answered 80% of CalendarQA date questions correctly; no other model of seven passed 47%
    Source: Saxena et al., 2025
    Chart: https://aiadoption.org/data/calendarqa-date-reasoning/
clockbench-leaderboard.csv  The best AI model read 50.6% of ClockBench clocks correctly in March 2026; humans manage about 90%
    Source: ClockBench Leaderboard, 2026
    Chart: https://aiadoption.org/data/clockbench-leaderboard/
mvbench-video-understanding-leaderboard.csv  JT3.5 and JT-VL-Chat tie for first on the MVBench video test at 74.1%; the top 15 span 23 points
    Source: MVBench Leaderboard, 2026
    Chart: https://aiadoption.org/data/mvbench-video-understanding-leaderboard/
video-mmmu-leaderboard.csv  No model reached the 74.4% human baseline on Video-MMMU; the best, Keye-VL-1.5-8B, scored 66%
    Source: Video-MMMU Leaderboard, 2026
    Chart: https://aiadoption.org/data/video-mmmu-leaderboard/
video-mmmu-knowledge-gain.csv  On Video-MMMU's measure of learning from video, human experts gain 33.1% and the best model, GPT-4o, 15.6%
    Source: Video-MMMU Leaderboard, 2026
    Chart: https://aiadoption.org/data/video-mmmu-knowledge-gain/
video-bench-video-quality.csv  Gen3 scores highest on all four Video-Bench quality measures, and motion is every model's weakest
    Source: Video-Bench Leaderboard, 2025
    Chart: https://aiadoption.org/data/video-bench-video-quality/
vbench-2-video-generation-leaderboard.csv  Veo 3 leads the VBench-2.0 video generation test at 66.7%, 4.0 points ahead of Vidu Q1
    Source: VBench Leaderboard, 2026
    Chart: https://aiadoption.org/data/vbench-2-video-generation-leaderboard/
taxeval-ai-tax-accuracy.csv  The top 15 models on TaxEval v2 scored within 2.9 percentage points of each other in early 2026
    Source: Vals.ai, 2026
    Chart: https://aiadoption.org/data/taxeval-ai-tax-accuracy/
mortgagetax-document-extraction.csv  No model reached 70% on MortgageTax in early 2026; Gemini 3.1 Pro Preview led at 69.4%
    Source: Vals.ai, 2026
    Chart: https://aiadoption.org/data/mortgagetax-document-extraction/
corpfin-credit-agreement-accuracy.csv  Kimi K2.5 led CorpFin v2's credit-agreement questions at 68.3% in early 2026; none reached 70%
    Source: Vals.ai, 2026
    Chart: https://aiadoption.org/data/corpfin-credit-agreement-accuracy/
finance-agent-benchmark.csv  Claude Sonnet 4.6 led Finance Agent at 63.3% in early 2026; no model reached two-thirds
    Source: Vals.ai, 2026
    Chart: https://aiadoption.org/data/finance-agent-benchmark/
caselaw-legal-research-accuracy.csv  GPT 5.1 led CaseLaw v2 legal research at 73.4% in early 2026, 3.5 points ahead of GPT 4.1
    Source: Vals.ai, 2026
    Chart: https://aiadoption.org/data/caselaw-legal-research-accuracy/
legalbench-legal-reasoning.csv  All top 15 models scored above 83% on LegalBench in early 2026; Gemini 3.1 Pro Preview led at 87.4%
    Source: Vals.ai, 2026
    Chart: https://aiadoption.org/data/legalbench-legal-reasoning/
