page_url,page_updated,theme,metric_id,metric,unit,year,value,value_numeric,as_of,source,source_url,confidence,note https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,gpt35_bar_tokens_per_dollar,"Tokens per $1 at GPT-3.5 quality (Stanford HAI series, MMLU 64.8)",tokens per US dollar (3:1 blend where input and output differ),2022,"50,000",50000,Nov 2022,"Stanford HAI, AI Index 2025: GPT-3.5-level inference at $20.00 per 1M tokens (Nov 2022)",https://hai.stanford.edu/ai-index/2025-ai-index-report,DERIVED,"1,000,000 / 20 = 50,000. Same as OpenAI's text-davinci-003 price." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,gpt35_bar_tokens_per_dollar,"Tokens per $1 at GPT-3.5 quality (Stanford HAI series, MMLU 64.8)",tokens per US dollar (3:1 blend where input and output differ),2023,"800,000",800000,Nov 2023,"OpenAI, gpt-3.5-turbo-1106 list price $1 input / $2 output per 1M (DevDay, 6 Nov 2023)",https://openai.com/index/new-models-and-developer-products-announced-at-devday/,DERIVED,"(3 x 1 + 2) / 4 = $1.25 per 1M. In March 2023 gpt-3.5-turbo launched at $2 per 1M (500,000 per $1), a tenth of earlier GPT-3.5 models. Hosted open models of similar quality (e.g. Mixtral 8x7B, Dec 2023) may have been cheaper on some hosts; this is the official list price, so a floor on tokens per dollar. Stanford's own Aug 2023 point is shown on page 421 as illustrative and is not used." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,gpt35_bar_tokens_per_dollar,"Tokens per $1 at GPT-3.5 quality (Stanford HAI series, MMLU 64.8)",tokens per US dollar (3:1 blend where input and output differ),2024,14.3M,14285714,Oct 2024,"Stanford HAI, AI Index 2025: $0.07 per 1M tokens (Oct 2024)",https://hai.stanford.edu/ai-index/2025-ai-index-report,DERIVED,"1,000,000 / 0.07 = 14.3M. 2022 to 2024: about 286x (Stanford's 'more than 280x')." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,gpt35_bar_tokens_per_dollar,"Tokens per $1 at GPT-3.5 quality (Stanford HAI series, MMLU 64.8)",tokens per US dollar (3:1 blend where input and output differ),2025,,,,,,,No sourced update of Stanford's GPT-3.5-level series for 2025 was found; GPT-3.5 quality is now below the GPT-4 bar tracked in gpt4_bar_tokens_per_dollar. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,gpt35_bar_tokens_per_dollar,"Tokens per $1 at GPT-3.5 quality (Stanford HAI series, MMLU 64.8)",tokens per US dollar (3:1 blend where input and output differ),2026,,,,,,,"No sourced 2026 reading. At the higher GPT-4 bar $1 bought about 5.0M tokens in Oct 2026, so GPT-3.5 quality is at least that cheap, but the cheapest GPT-3.5-level price is not published." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,gpt4_bar_tokens_per_dollar,"Tokens per $1 at GPT-4 (March 2023) quality, cheapest named model",tokens per US dollar (3:1 input:output blend),2022,,,,,,,No model at GPT-4 level existed; GPT-4 was released on 14 Mar 2023. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,gpt4_bar_tokens_per_dollar,"Tokens per $1 at GPT-4 (March 2023) quality, cheapest named model",tokens per US dollar (3:1 input:output blend),2023,"66,667",66667,Nov 2023,"OpenAI, GPT-4 Turbo (gpt-4-1106-preview) list price $10 input / $30 output per 1M, DevDay 6 Nov 2023",https://openai.com/index/new-models-and-developer-products-announced-at-devday/,DERIVED,"(3 x 10 + 30) / 4 = $15.00 per 1M; 1,000,000 / 15 = 66,667. GPT-4 Turbo was the cheapest model at the GPT-4 bar by year end. At GPT-4's launch (14 Mar 2023, $30 / $60) the blend was $37.50 = 26,667 tokens per $1; the at-a-glance and chapter 1 '188x' comparisons use that launch figure." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,gpt4_bar_tokens_per_dollar,"Tokens per $1 at GPT-4 (March 2023) quality, cheapest named model",tokens per US dollar (3:1 input:output blend),2024,2.09M,2094241,Dec 2024,DeepSeek API list price (DeepSeek-V3: $0.27 input / $1.10 output per 1M); Report AI calculation,https://api-docs.deepseek.com/quick_start/pricing,DERIVED,"Blended price = (3 x input + output) / 4 = (3 x 0.27 + 1.10) / 4 = $0.4775 per 1M tokens; tokens per $1 = 1,000,000 / 0.4775. Bar: published MMLU at or above GPT-4's 86.4 (DeepSeek-V3 reports 88.5). A promotional price of $0.14 / $0.28 ran until 8 Feb 2025 and is not used." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,gpt4_bar_tokens_per_dollar,"Tokens per $1 at GPT-4 (March 2023) quality, cheapest named model",tokens per US dollar (3:1 input:output blend),2025,3.17M,3174603,Sep 2025,DeepSeek API list price (DeepSeek-V3.2-Exp: $0.28 input cache-miss / $0.42 output per 1M); Report AI calculation,https://api-docs.deepseek.com/quick_start/pricing,DERIVED,"(3 x 0.28 + 0.42) / 4 = $0.315 per 1M. Smaller models that clear GPT-4 on some benchmarks but have no published MMLU above 86.4 (e.g. GPT-5 nano at $0.05 / $0.40, = 7.27M tokens per $1) are excluded, so this series is a conservative floor on tokens per dollar." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,gpt4_bar_tokens_per_dollar,"Tokens per $1 at GPT-4 (March 2023) quality, cheapest named model",tokens per US dollar (3:1 input:output blend),2026,5.0M,5000000,Oct 2026 (to date),"OpenAI GPT-6 Luna ($0.10 / $0.50, after the 22 Sep 2026 cut) and Anthropic Claude Haiku 5.5 ($0.10 / $0.50 for prompts up to 100K tokens, 7 Oct 2026), as reported; Report AI calculation",https://www.techrepublic.com/article/news-anthropic-claude-haiku-5-5-api-pricing/,DERIVED,"(3 x 0.10 + 0.50) / 4 = $0.20 per 1M. Prices are MEDIUM (secondary reporting; vendor pages not opened). Anthropic says Haiku 5.5's tokenizer uses about 30% more tokens for the same text, so tokens per dollar overstates text per dollar for that model. gpt-oss-120b listed at $0.09 / $0.45 on an aggregator in Aug 2026 (blended $0.18, 5.6M per $1; single source, not used). Change from GPT-4's March 2023 launch to 2026: 26,667 to 5,000,000 = about 188x; from the 2023 year-end value (66,667): 75x; 2024 to 2026 = about 2.4x." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openai_flagship_tokens_per_dollar,Tokens per $1 at OpenAI's flagship list price,tokens per US dollar (3:1 blend),2022,"50,000",50000,Sep-Dec 2022,"OpenAI pricing: Davinci (GPT-3) cut from $0.06 to $0.02 per 1,000 tokens on 1 Sep 2022; text-davinci-003 (28 Nov 2022) at the same $0.02",https://openai.com/api/pricing/,DERIVED,"$20 per 1M for input and output alike, so the 3:1 blend is $20.00; 1,000,000 / 20 = 50,000. Price history from OpenAI's 2022 announcements via secondary reporting (MEDIUM)." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openai_flagship_tokens_per_dollar,Tokens per $1 at OpenAI's flagship list price,tokens per US dollar (3:1 blend),2023,"66,667",66667,Nov 2023,"OpenAI, GPT-4 Turbo list price $10 / $30 per 1M (DevDay, 6 Nov 2023)",https://openai.com/index/new-models-and-developer-products-announced-at-devday/,DERIVED,"(3 x 10 + 30) / 4 = $15.00. GPT-4 at launch (Mar 2023, $30 / $60) = $37.50, 26,667 tokens per $1, fewer than GPT-3 Davinci bought in 2022: the new frontier was priced at a premium." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openai_flagship_tokens_per_dollar,Tokens per $1 at OpenAI's flagship list price,tokens per US dollar (3:1 blend),2024,"228,571",228571,Aug 2024,"OpenAI API pricing, GPT-4o ($2.50 input / $10.00 output per 1M)",https://openai.com/api/pricing/,DERIVED,(3 x 2.50 + 10) / 4 = $4.375 per 1M. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openai_flagship_tokens_per_dollar,Tokens per $1 at OpenAI's flagship list price,tokens per US dollar (3:1 blend),2025,"290,909",290909,Aug 2025,"OpenAI API pricing, GPT-5 ($1.25 input / $10.00 output per 1M)",https://openai.com/api/pricing/,DERIVED,(3 x 1.25 + 10) / 4 = $3.4375 per 1M. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openai_flagship_tokens_per_dollar,Tokens per $1 at OpenAI's flagship list price,tokens per US dollar (3:1 blend),2026,"250,000",250000,Oct 2026 (to date),"GPT-6.1 Sol at $2 / $10 (implied by GPT-6.1 Terra at $1 / $5 being half its price, 26 Oct 2026; GPT-6 Sol cut 50% to $2 / $10 on 22 Sep 2026), as reported by Technobezz and iClarified",https://www.technobezz.com/news/openai-gpt-6-1-terra-cuts-api-prices,DERIVED,"(3 x 2 + 10) / 4 = $4.00 per 1M. Inputs MEDIUM; OpenAI pricing page not checked. Earlier in 2026 the flagship GPT-5.6 Sol listed at $5 / $30 (Jul 2026, page 964): blended $11.25, = 88,889 tokens per $1, so the 2026 figure moved 2.8x within the year. Flagship tokens per dollar rose about 9x from 2023 to 2024 and has been roughly flat since." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,anthropic_flagship_tokens_per_dollar,Tokens per $1 at Anthropic's flagship list price (Opus tier from 2024),tokens per US dollar (3:1 blend),2022,,,,,,,Anthropic had no generally available paid API in 2022. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,anthropic_flagship_tokens_per_dollar,Tokens per $1 at Anthropic's flagship list price (Opus tier from 2024),tokens per US dollar (3:1 blend),2023,"83,333",83333,Jul-Nov 2023,"Anthropic pricing, Claude 2 and Claude 2.1 ($8 input / $24 output per 1M)",https://www.anthropic.com/news/claude-2,DERIVED,"(3 x 8 + 24) / 4 = $12.00. Claude 2 was Anthropic's flagship but not an 'Opus' tier; Claude 3 Opus (2024) was priced higher, which is why the series dips in 2024. Claude Instant 1.2 listed at $1.63 / $5.51 (blended $2.60, 384,615 per $1)." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,anthropic_flagship_tokens_per_dollar,Tokens per $1 at Anthropic's flagship list price (Opus tier from 2024),tokens per US dollar (3:1 blend),2024,"33,333",33333,Mar 2024,"Anthropic pricing, Claude 3 Opus ($15 input / $75 output per 1M)",https://www.anthropic.com/pricing,DERIVED,(3 x 15 + 75) / 4 = $30.00 per 1M. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,anthropic_flagship_tokens_per_dollar,Tokens per $1 at Anthropic's flagship list price (Opus tier from 2024),tokens per US dollar (3:1 blend),2025,"100,000",100000,Nov 2025,"Anthropic, Claude Opus 4.5 launch pricing ($5 / $25 per 1M)",https://www.anthropic.com/news/claude-opus-4-5,DERIVED,"(3 x 5 + 25) / 4 = $10.00 per 1M. Opus 4 and 4.1 listed at $15 / $75 for most of 2025 (33,333 per $1)." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,anthropic_flagship_tokens_per_dollar,Tokens per $1 at Anthropic's flagship list price (Opus tier from 2024),tokens per US dollar (3:1 blend),2026,"125,000",125000,Sep 2026 (to date),"Anthropic pricing page, Claude Opus 5.5 ($4 / $20 per 1M), verified by Report AI on 1 Oct 2026",https://www.anthropic.com/pricing,DERIVED,"(3 x 4 + 20) / 4 = $8.00 per 1M. Inputs HIGH (page 421 verification). Anthropic's top tier, Claude Fable 5.1 at $10 / $50, blends to $20.00 = 50,000 tokens per $1." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,usage_weighted_market_price,"Usage-weighted market price per million tokens (Silicon Data LLM Token Expenditure Index, SDLLMTK)",US$ per 1M tokens,2022,,,,,,,Index did not exist. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,usage_weighted_market_price,"Usage-weighted market price per million tokens (Silicon Data LLM Token Expenditure Index, SDLLMTK)",US$ per 1M tokens,2023,,,,,,,Index did not exist. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,usage_weighted_market_price,"Usage-weighted market price per million tokens (Silicon Data LLM Token Expenditure Index, SDLLMTK)",US$ per 1M tokens,2024,,,,,,,Index did not exist; no usage-weighted market price series for 2024 was found. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,usage_weighted_market_price,"Usage-weighted market price per million tokens (Silicon Data LLM Token Expenditure Index, SDLLMTK)",US$ per 1M tokens,2025,~$1.09,1.09,Dec 2025 (launch),Implied by a third-party explainer saying the index nearly doubled from about 1.09 to about 2.12 between December 2025 and May 2026,https://sanj.dev/post/sdllmtk-llm-token-expenditure-index/,MEDIUM,Single secondary source; Silicon Data's launch value not seen. Treat as approximate. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,usage_weighted_market_price,"Usage-weighted market price per million tokens (Silicon Data LLM Token Expenditure Index, SDLLMTK)",US$ per 1M tokens,2026,~$0.95,0.95,22 Oct 2026 (to date),"Axios, citing Silicon Data: index fell below $1, 54% under its late-May peak of $2.07",https://axios.com/2026/10/22/ai-token-prices-demand,DERIVED,"2.07 x (1 - 0.54) = 0.95 (same calculation as page 3441). 2026 readings found: 22 Apr frontier segment $4.20 and open-weight segment $0.85 (Silicon Data product page); peak ~$2.07 on about 28 May (Axios, Tokenpost; others give $2.12 or $2.05, CONTESTED); ~$1.62 in early July, about 20% under the peak (Bloomberg via secondary outlets); $1.16-1.18 on 6-8 Aug (SCMP); $0.97 on 31 Aug / 1 Sep (CNBC, Enterprise DNA); ~$1.00 on 21 Sep (Tokenpost). The 1 Sep reading was already below $1, so Axios's 'recent' break below $1 on 22 Oct implies a rebound in between or a lag (CONTESTED, as on page 3441). The index blends price changes with buyers switching models; it is not a list-price index and not a volume index." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,fixed_capability_price_decline,Annual decline in the price of a fixed level of capability,times cheaper per year,2022,,,,,,,No single-year reading; Stanford's 280x decline spans Nov 2022 to Oct 2024 (see gpt35_bar_tokens_per_dollar). https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,fixed_capability_price_decline,Annual decline in the price of a fixed level of capability,times cheaper per year,2023,,,,,,,No single-year reading found. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,fixed_capability_price_decline,Annual decline in the price of a fixed level of capability,times cheaper per year,2024,~10x per year,10,Nov 2024,"a16z, Welcome to LLMflation (GPT-3-level MMLU 42: $60 per 1M in Nov 2021 to $0.06 in Nov 2024, 1,000x in three years)",https://a16z.com/llmflation-llm-inference-cost/,MEDIUM,"Stanford HAI AI Index 2025: GPT-3.5-level (MMLU 64.8) fell from $20.00 per 1M (Nov 2022) to $0.07 (Oct 2024), more than 280x (page 421)." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,fixed_capability_price_decline,Annual decline in the price of a fixed level of capability,times cheaper per year,2025,~50x per year (median; range 9x-900x),50,Mar 2025,"Epoch AI, LLM inference price trends",https://epoch.ai/blog/llm-inference-price-trends,MEDIUM,Epoch found the price of GPT-4-level performance on PhD-level science questions (GPQA) fell about 40x per year. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,fixed_capability_price_decline,Annual decline in the price of a fixed level of capability,times cheaper per year,2026,~13x per year,12.7,2026 (to date),"Epoch AI, The plunging price of thought: cost of a fixed performance level fell about 47% per quarter over three years",https://epoch.ai/publications/the-plunging-price-of-thought,DERIVED,"1 / (1 - 0.47)^4 = 12.7x per year. Epoch calls its figures rough; five benchmarks only. Not like-for-like with the 2024 and 2025 readings (different methods), so yoy is false." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,frontier_benchmark_cost,Cost of running a standard benchmark on a frontier reasoning model,"US$ (see note: suite total in 2025, per task in 2026)",2022,,,,,,,Not measured. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,frontier_benchmark_cost,Cost of running a standard benchmark on a frontier reasoning model,"US$ (see note: suite total in 2025, per task in 2026)",2023,,,,,,,Not measured. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,frontier_benchmark_cost,Cost of running a standard benchmark on a frontier reasoning model,"US$ (see note: suite total in 2025, per task in 2026)",2024,,,,,,,No comparable reading found. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,frontier_benchmark_cost,Cost of running a standard benchmark on a frontier reasoning model,"US$ (see note: suite total in 2025, per task in 2026)",2025,"$2,767 (o1, seven-benchmark suite) vs $109 (GPT-4o)",2767,Apr 2025,Artificial Analysis via TechCrunch / DeepLearning.AI The Batch,https://www.deeplearning.ai/the-batch/reasoning-llms-are-pricey-to-test,MEDIUM,"The reasoning model cost about 25x more to evaluate than the non-reasoning model on the same suite, mostly because it generated far more tokens." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,frontier_benchmark_cost,Cost of running a standard benchmark on a frontier reasoning model,"US$ (see note: suite total in 2025, per task in 2026)",2026,"$10.04 per task (Claude Opus 5.5, max) vs $1.27 (GPT-6.1 Sol, max)",10.04,28 Oct 2026 (to date),"Artificial Analysis Intelligence Index v5, as reported by FourWeekMBA",https://fourweekmba.com/ai-opus-5-5-tops-ai-index-at-58-gpt-6-1-sol-costs-8x-less/,MEDIUM,"Single secondary source. Opus 5.5 scored 58 and GPT-6.1 Sol 55; a 7.9x cost-per-task gap against a 2x gap in list price per token, because models differ in how many tokens they spend per task. Not comparable with the 2025 suite total." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,google_monthly_tokens,Google: tokens processed per month across all products,tokens per month,2022,,,,,,,Not disclosed. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,google_monthly_tokens,Google: tokens processed per month across all products,tokens per month,2023,,,,,,,Not disclosed; Google's first published monthly figure (9.7T) refers to spring 2024. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,google_monthly_tokens,Google: tokens processed per month across all products,tokens per month,2024,9.7T,9700000000000,Apr-May 2024,"Alphabet (Sundar Pichai, Google I/O, May 2025, citing the year-earlier figure)",https://www.tomtunguz.com/trillion-token-race/index.md,MEDIUM,Only 2024 reading disclosed. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,google_monthly_tokens,Google: tokens processed per month across all products,tokens per month,2025,1.3 quadrillion,1300000000000000,Oct 2025,Alphabet Q3 2025 earnings remarks,https://www.tomtunguz.com/is-token-consumption-slowing-down/index.md,MEDIUM,"Earlier 2025 readings: 480T (May 2025), 980T (Jul 2025). May-to-May growth 2024 to 2025: about 49x." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,google_monthly_tokens,Google: tokens processed per month across all products,tokens per month,2026,3.2 quadrillion,3200000000000000,May 2026 (latest all-surface figure),"Alphabet, Google I/O 2026",https://gigazine.net/gsc_news/en/20260520-google-monthly-tokens-processed/,MEDIUM,"Google presented it as about 7x year on year (480T in May 2025). Self-reported, unaudited. No all-surface figure was given at the Q2 or Q3 2026 earnings calls, which reported API throughput only (see google_api_tokens_per_minute)." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,google_api_tokens_per_minute,Google: first-party model API tokens per minute,tokens per minute,2022,,,,,,,Not disclosed. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,google_api_tokens_per_minute,Google: first-party model API tokens per minute,tokens per minute,2023,,,,,,,Not disclosed. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,google_api_tokens_per_minute,Google: first-party model API tokens per minute,tokens per minute,2024,,,,,,,Not disclosed. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,google_api_tokens_per_minute,Google: first-party model API tokens per minute,tokens per minute,2025,10B+,10000000000,Q4 2025,Alphabet Q4 2025 earnings remarks (series as tabulated by PPC Land),https://ppc.land/gemini-app-gains-350m-users-in-a-year-to-pass-1bn-alphabet-says/,MEDIUM,Q3 2025: about 7B per minute. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,google_api_tokens_per_minute,Google: first-party model API tokens per minute,tokens per minute,2026,~32B,32000000000,"Q3 2026 (28 Oct 2026, to date)","Alphabet Q3 2026 earnings remarks: nearly 32 billion tokens per minute, 40% more than last quarter and more than 4x a year ago",https://blog.google/company-news/inside-google/message-ceo/alphabet-earnings-q3-2026/,MEDIUM,"Q1 2026: 16B; Q2 2026: 22B. At 32B per minute a 30-day month is about 1.38 quadrillion API tokens (PPC Land's calculation; Report AI agrees: 32e9 x 43,200 = 1.38e15). API only, not all surfaces." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openai_api_tokens_per_minute,OpenAI: API tokens per minute,tokens per minute,2022,,,,,,,"Not disclosed. ChatGPT launched on 30 Nov 2022; usage was reported in users, not tokens." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openai_api_tokens_per_minute,OpenAI: API tokens per minute,tokens per minute,2023,~300M,300000000,"2023 (DevDay, Nov 2023 era)","OpenAI, Sam Altman at DevDay 2025, recalling the 2023 figure",https://openai.com/devday/,MEDIUM,"About 432 billion tokens a day (300M x 1,440). Only token-volume disclosure found for 2023." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openai_api_tokens_per_minute,OpenAI: API tokens per minute,tokens per minute,2024,,,,,,,No 2024 figure found. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openai_api_tokens_per_minute,OpenAI: API tokens per minute,tokens per minute,2025,6B+,6000000000,6 Oct 2025,"OpenAI, DevDay 2025 (Sam Altman keynote)",https://openai.com/devday/,MEDIUM,"Equals about 8.6 trillion tokens a day, matching a16z's 'about 8.6 trillion per day in October' comparison." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openai_api_tokens_per_minute,OpenAI: API tokens per minute,tokens per minute,2026,,,,,,,DevDay 2026 (29 Sep 2026) gave no throughput figure. A '15B+ per minute' figure appears on an unattributed profile page and is not used. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,microsoft_tokens_per_quarter,Microsoft: tokens processed per quarter (company-wide),tokens per quarter,2022,,,,,,,Not disclosed. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,microsoft_tokens_per_quarter,Microsoft: tokens processed per quarter (company-wide),tokens per quarter,2023,,,,,,,Not disclosed. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,microsoft_tokens_per_quarter,Microsoft: tokens processed per quarter (company-wide),tokens per quarter,2024,~20T,20000000000000,Jan-Mar 2024,"Microsoft FY25 Q3 call ('over 100 trillion tokens this quarter, up 5x year over year'); Report AI calculation",https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/TranscriptFY25Q3,DERIVED,100T / 5 = 20T. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,microsoft_tokens_per_quarter,Microsoft: tokens processed per quarter (company-wide),tokens per quarter,2025,100T+,100000000000000,Jan-Mar 2025,Microsoft FY25 Q3 earnings call,https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/TranscriptFY25Q3,MEDIUM,"50T of it in March 2025 alone. Separately, Foundry APIs served more than 500T tokens in fiscal 2025 (Jul 2024 to Jun 2025), up more than 7x (FY25 Q4 call)." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,microsoft_tokens_per_quarter,Microsoft: tokens processed per quarter (company-wide),tokens per quarter,2026,,,,,,,"No total disclosed. Microsoft now reports customers: more than 300 on track to process over 1 trillion tokens each on Foundry this year (FY26 Q3 call, Apr 2026; 250 in Jan 2026)." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openrouter_weekly_tokens,OpenRouter: tokens routed per week,tokens per week,2022,,,,,,,OpenRouter did not exist (founded 2023). https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openrouter_weekly_tokens,OpenRouter: tokens routed per week,tokens per week,2023,,,,,,,No 2023 volume published. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openrouter_weekly_tokens,OpenRouter: tokens routed per week,tokens per week,2024,~10B,10000000000,Jan 2024,"Crypto Briefing, citing OpenRouter's growth chart (9,000x since January 2024)",https://cryptobriefing.com/openrouter-9000x-token-usage-growth/,MEDIUM,Single secondary source for the baseline. The Decoder gives about 0.5T per week in January 2025. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openrouter_weekly_tokens,OpenRouter: tokens routed per week,tokens per week,2025,~7T,7000000000000,Dec 2025,"a16z / OpenRouter, State of AI: An Empirical 100 Trillion Token Study (more than 1 trillion tokens every day in the week before publication)",https://a16z.com/state-of-ai/,DERIVED,1T per day x 7 = 7T per week (floor). An OpenRouter engineer's chart shows 4.6T per week in September 2025. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openrouter_weekly_tokens,OpenRouter: tokens routed per week,tokens per week,2026,~194T,194000000000000,week ending 26 Oct 2026 (to date),"Chart posted by Menlo Ventures (an OpenRouter investor), as reported by FourWeekMBA and Crypto Briefing",https://fourweekmba.com/ai-openrouters-run-rate-tops-10-quadrillion-tokens-up-6-7x/,MEDIUM,"Not published by OpenRouter itself. OpenRouter's own October 2026 recap: 790T tokens in the month, up 37% on September and 25x January (paid traffic only). Other 2026 weekly readings: 6.4T (first week of January), 13T (week to 9 Feb), 126-128T (mid-September)." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,doubao_daily_tokens,ByteDance Doubao models: tokens processed per day,tokens per day,2022,,,,,,,Not launched (Doubao models launched May 2024). https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,doubao_daily_tokens,ByteDance Doubao models: tokens processed per day,tokens per day,2023,,,,,,,Not launched (Doubao models launched May 2024). https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,doubao_daily_tokens,ByteDance Doubao models: tokens processed per day,tokens per day,2024,4T+,4000000000000,Dec 2024,"Volcano Engine, Winter FORCE conference, as reported by SCMP",https://www.scmp.com/tech/tech-trends/article/3336895/bytedance-ai-model-usage-grows-over-tenfold-amid-increased-chinese-enterprise-adoption,MEDIUM,33x the level at the May 2024 launch. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,doubao_daily_tokens,ByteDance Doubao models: tokens processed per day,tokens per day,2025,~50T,50000000000000,18 Dec 2025,"Volcano Engine president Tan Dai, FORCE conference",https://www.scmp.com/tech/tech-trends/article/3336895/bytedance-ai-model-usage-grows-over-tenfold-amid-increased-chinese-enterprise-adoption,MEDIUM,Mid-2025: 16.4T (May 2025). Reports call December 2025 'more than ten times' a year earlier; the multiple depends on the baseline used. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,doubao_daily_tokens,ByteDance Doubao models: tokens processed per day,tokens per day,2026,180T+,180000000000000,Jun 2026 (latest found),"Volcano Engine, FORCE conference (Doubao 2.1 launch), as reported by TMTPost and others",https://www.kucoin.com/news/flash/volcanic-engine-launches-doubao-2-1-with-daily-token-usage-exceeding-180-trillion,MEDIUM,"Includes internal ByteDance use and the Doubao app as well as external API customers. March 2026: 120T (TechNode, China Daily). At 180T per day a 30-day month is 5.4 quadrillion, larger than Google's 3.2 quadrillion all-surface figure for May 2026; definitions differ and neither is audited." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,china_daily_tokens,"China: average daily token calls, national",tokens per day,2022,,,,,,,Not published. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,china_daily_tokens,"China: average daily token calls, national",tokens per day,2023,,,,,,,Not published; the NDA series starts in early 2024. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,china_daily_tokens,"China: average daily token calls, national",tokens per day,2024,~0.1T (100B),100000000000,early 2024,"National Data Administration (NDA), as reported by CnTechPost and China Daily",https://cntechpost.com/2026/03/24/china-daily-ai-token-usage-140-trillion,CONTESTED,NDA's series starts at about 100 billion in early 2024; remarks at the Digital China Summit instead put the start of 2025 at 'over 1 trillion'. Both are official; the baselines differ. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,china_daily_tokens,"China: average daily token calls, national",tokens per day,2025,~100T,100000000000000,end 2025,"NDA, National Data Resource Survey Report (2025)",https://cntechpost.com/2026/03/24/china-daily-ai-token-usage-140-trillion,MEDIUM,"Annual 2025 total reported at 21.1 quadrillion tokens (CnTechPost, 29 Apr 2026)." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,china_daily_tokens,"China: average daily token calls, national",tokens per day,2026,140T+,140000000000000,Mar 2026 (latest dated figure),"NDA director Liu Liehong, policy briefing",https://cntechpost.com/2026/03/24/china-daily-ai-token-usage-140-trillion,MEDIUM,"An undated MIIT-affiliated report cites more than 175T per day; not used because its date is unknown. IDC expects China's 2026 token consumption to be about 20x 2025's (forecast, not an outcome)." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openrouter_agent_token_share,Share of OpenRouter tokens consumed by AI agents,% of tokens,2022,,,,,,,Not measured. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openrouter_agent_token_share,Share of OpenRouter tokens consumed by AI agents,% of tokens,2023,,,,,,,Not measured. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openrouter_agent_token_share,Share of OpenRouter tokens consumed by AI agents,% of tokens,2024,,,,,,,Not measured. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openrouter_agent_token_share,Share of OpenRouter tokens consumed by AI agents,% of tokens,2025,~8%,8,Oct 2025,Secondary write-up of OpenRouter's October 2026 recap ('8% a year ago for agentic keys'),https://fourweekmba.com/ai-openrouter-agents-used-76-of-its-790t-tokens-in-october/,MEDIUM,Single secondary source; definition (agentic API keys) set by OpenRouter. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openrouter_agent_token_share,Share of OpenRouter tokens consumed by AI agents,% of tokens,2026,76%,76,Oct 2026 (to date),"OpenRouter, October 2026 data recap",https://openrouter.ai/data/october-2026,MEDIUM,"About 600T of October's 790T tokens. About 71% in early August 2026 (Crypto Briefing). One analyst summary says 75%. Menlo's November 2026 report cites 80%, also from OpenRouter data." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,disclosed_monthly_tokens_google_bytedance,"Disclosed tokens per month, Google plus ByteDance (cross-platform floor)",tokens per month,2022,,,,,,,Inputs not disclosed. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,disclosed_monthly_tokens_google_bytedance,"Disclosed tokens per month, Google plus ByteDance (cross-platform floor)",tokens per month,2023,,,,,,,Inputs not disclosed. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,disclosed_monthly_tokens_google_bytedance,"Disclosed tokens per month, Google plus ByteDance (cross-platform floor)",tokens per month,2024,~130T,129700000000000,2024 (May and Dec readings),Report AI calculation from Alphabet and Volcano Engine disclosures,,DERIVED,"Google 9.7T per month (Apr-May 2024) + Doubao 4T per day x 30 = 120T (Dec 2024). Normalised to a 30-day month. Excludes OpenAI, Anthropic, Microsoft, Alibaba, Meta, AWS and all self-hosted models." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,disclosed_monthly_tokens_google_bytedance,"Disclosed tokens per month, Google plus ByteDance (cross-platform floor)",tokens per month,2025,~2.8 quadrillion,2800000000000000,Oct-Dec 2025,Report AI calculation,,DERIVED,"Google 1.3 quadrillion (Oct 2025) + Doubao 50T x 30 = 1.5 quadrillion (Dec 2025). OpenAI's API at 6B per minute (Oct 2025) would add about 0.26 quadrillion but has no 2024 or 2026 equivalent, so it is left out to keep the series consistent. OpenRouter is excluded because it routes to the same lab APIs (double counting); China's national figure is excluded because it contains Doubao." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,disclosed_monthly_tokens_google_bytedance,"Disclosed tokens per month, Google plus ByteDance (cross-platform floor)",tokens per month,2026,~8.6 quadrillion,8600000000000000,May-Jun 2026 (to date),Report AI calculation,,DERIVED,"Google 3.2 quadrillion (May 2026) + Doubao 180T x 30 = 5.4 quadrillion (Jun 2026). A floor, not a market total: the two firms count differently (Doubao includes its own consumer app; Google counts all its products) and most providers disclose nothing comparable." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,enterprise_genai_spend,Enterprise generative-AI spending (Menlo Ventures estimate; definition changed in 2026),US$ per year,2022,,,,,,,Menlo's estimates start in 2023. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,enterprise_genai_spend,Enterprise generative-AI spending (Menlo Ventures estimate; definition changed in 2026),US$ per year,2023,$1.7B,1700000000,2023,"Menlo Ventures, 2025 report footnote (2023 estimate excluding inference); originally $2.3B in its Nov 2024 report",https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/,MEDIUM,"RESTATED: $2.3B in Menlo's 2024 report (which included inference) and $1.7B on the 2025 report's basis, the same basis as $11.5B for 2024. Both kept. The 2026 figure uses a new, non-comparable definition, so no multiple to 2026 is published." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,enterprise_genai_spend,Enterprise generative-AI spending (Menlo Ventures estimate; definition changed in 2026),US$ per year,2024,$11.5B,11500000000,2024,"Menlo Ventures, 2025: The State of Generative AI in the Enterprise",https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/,MEDIUM,"Menlo's footnote says its 2024 estimate excludes inference that its earlier reports had included, so the series has a scope change." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,enterprise_genai_spend,Enterprise generative-AI spending (Menlo Ventures estimate; definition changed in 2026),US$ per year,2025,$37B,37000000000,2025,"Menlo Ventures, Dec 2025 report",https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/,MEDIUM,"RESTATED: Menlo's November 2026 report restates 2025 as $33B. Both numbers kept. Split in the Dec 2025 report: about $19B applications, $18B infrastructure (page 46)." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,enterprise_genai_spend,Enterprise generative-AI spending (Menlo Ventures estimate; definition changed in 2026),US$ per year,2026,$176.8B (annualised),176800000000,Sep 2026 (to date),"Menlo Ventures, 2026: The State of Enterprise AI (3 Nov 2026)",https://menlovc.com/wp-content/uploads/2026/11/menlo_ventures_enterprise_ai_report-2026-110326.pdf,CONTESTED,"A run rate on Menlo's NEW 2026 definition (enterprise AI deployment spending), not cash spent in 2026 and not comparable with the $11.5B (2024) or $37B / restated $33B (2025) annual totals; Report AI's State of AI Spending report rates the comparison 'not comparable'. Menlo itself presents it as 5.4x its restated 2025 figure; we do not publish a 2024-to-2026 or 2025-to-2026 multiple from this series. Components: $99.5B infrastructure and $77.3B applications; one analysis puts foundation models and inference at $94.2B. Menlo is an investor in Anthropic and OpenRouter. Report not opened; figures from consistent secondary coverage. yoy set to false for this reason." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,enterprise_llm_api_spend,Enterprise spending on LLM APIs (Menlo Ventures),US$ per year (annualised),2022,,,,,,,Not estimated. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,enterprise_llm_api_spend,Enterprise spending on LLM APIs (Menlo Ventures),US$ per year (annualised),2023,,,,,,,"No comparable Menlo 2023 dollar figure for LLM API spend found (one secondary summary cites about $1B of 2023 foundation-model spend; unverified, not used)." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,enterprise_llm_api_spend,Enterprise spending on LLM APIs (Menlo Ventures),US$ per year (annualised),2024,$3.5B,3500000000,Nov 2024,"Menlo Ventures, 2025 Mid-Year LLM Market Update",https://menlovc.com/insights/2025-mid-year-llm-market-update,MEDIUM, https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,enterprise_llm_api_spend,Enterprise spending on LLM APIs (Menlo Ventures),US$ per year (annualised),2025,$8.4B,8400000000,mid-2025,"Menlo Ventures, 2025 Mid-Year LLM Market Update",https://menlovc.com/insights/2025-mid-year-llm-market-update,MEDIUM,2.4x in about six months. Survey of 150 technical decision-makers. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,enterprise_llm_api_spend,Enterprise spending on LLM APIs (Menlo Ventures),US$ per year (annualised),2026,,,,,,,"Menlo's November 2026 report gives shares (Anthropic 55%, OpenAI 31%, Google 9% of enterprise LLM API spend, per secondary analyses) but no comparable dollar total for API spend." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openai_anthropic_run_rate,OpenAI plus Anthropic revenue (combined annualised run rate; 2022 recognised revenue),US$ per year (annualised),2022,"~$28M (OpenAI recognised revenue, full year)",28000000,2022,OpenAI recognised revenue for 2022 as reported in press coverage of its financials; Anthropic not disclosed,https://epoch.ai/data-insights/ai-companies-revenue,MEDIUM,"Different basis: full-year recognised revenue, not an annualised run rate, and OpenAI only. Shown for scale; not used in any index or multiple." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openai_anthropic_run_rate,OpenAI plus Anthropic revenue (combined annualised run rate; 2022 recognised revenue),US$ per year (annualised),2023,~$1.7B,1700000000,Oct-Dec 2023,"OpenAI annualised revenue passed $1.6B (The Information, via SiliconANGLE, 1 Jan 2024) + Anthropic about $100M (Epoch AI, dated 3 Oct 2023)",https://siliconangle.com/2024/01/01/openais-annualized-revenue-reportedly-tops-1-6b/,DERIVED,"Sum of two MEDIUM run-rate figures, two months apart. Anthropic counts cloud-partner sales in its run rate and OpenAI does not (FT, as reported), so the sum is approximate." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openai_anthropic_run_rate,OpenAI plus Anthropic revenue (combined annualised run rate; 2022 recognised revenue),US$ per year (annualised),2024,~$6.5B,6500000000,Dec 2024,"OpenAI about $5.5B (CNBC, Jun 2025, citing the year-earlier figure) + Anthropic about $1B (press reports)",,DERIVED,"Sum of two MEDIUM run-rate figures. Run rate = latest month x 12; includes consumer subscriptions, not only token sales." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openai_anthropic_run_rate,OpenAI plus Anthropic revenue (combined annualised run rate; 2022 recognised revenue),US$ per year (annualised),2025,~$29B,29000000000,end 2025,OpenAI about $20B (page 935) + Anthropic about $9B (Bloomberg via TheNextWeb),https://thenextweb.com/news/anthropic-revenue-run-rate-65-billion,DERIVED,Sum of two MEDIUM figures. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,openai_anthropic_run_rate,OpenAI plus Anthropic revenue (combined annualised run rate; 2022 recognised revenue),US$ per year (annualised),2026,~$115B,115000000000,Jul-Sep 2026 (to date),"OpenAI about $50B (end Sep 2026, CNBC) + Anthropic more than $65B (end Jul 2026, Bloomberg)",https://thenextweb.com/news/anthropic-revenue-run-rate-65-billion,DERIVED,"Dates differ by two months. Anthropic's S-1 reports H1 2026 revenue of $16.3B (recognised, not run rate), of which about 80% ($13.1B) was consumption-based 'pay for intelligence' revenue per one secondary reading. Index 2024 = 100: 446 (2025), 1,769 (2026); this is the revenue leg of the Jevons comparison (metric jevons_revenue_index)." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,jevons_volume_index,"Jevons check: token volume index (Google monthly tokens, May each year), 2024 = 100","index, 2024 = 100",2022,,,,,,,No Google volume disclosed. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,jevons_volume_index,"Jevons check: token volume index (Google monthly tokens, May each year), 2024 = 100","index, 2024 = 100",2023,,,,,,,"No Google volume disclosed; earliest is spring 2024, which is why the Jevons base year is 2024." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,jevons_volume_index,"Jevons check: token volume index (Google monthly tokens, May each year), 2024 = 100","index, 2024 = 100",2024,100,100,Apr-May 2024,Report AI calculation from Alphabet disclosures,,DERIVED,Base: 9.7T tokens per month. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,jevons_volume_index,"Jevons check: token volume index (Google monthly tokens, May each year), 2024 = 100","index, 2024 = 100",2025,"4,948",4948,May 2025,Report AI calculation,,DERIVED,"480T / 9.7T x 100. Same month each year, so seasonality cancels." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,jevons_volume_index,"Jevons check: token volume index (Google monthly tokens, May each year), 2024 = 100","index, 2024 = 100",2026,"32,990",32990,May 2026,Report AI calculation,,DERIVED,"3,200T / 9.7T x 100 (about 330x in two years). Google is one platform; Doubao (4T to 180T per day, Dec 2024 to Jun 2026, = 4,500) and OpenRouter grew on the same order." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,jevons_revenue_index,"Jevons check: revenue index, OpenAI + Anthropic run-rate revenue, 2024 = 100","index, 2024 = 100",2022,,,,,,,"2022 revenue is recognised revenue for OpenAI only, a different basis; not indexed." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,jevons_revenue_index,"Jevons check: revenue index, OpenAI + Anthropic run-rate revenue, 2024 = 100","index, 2024 = 100",2023,26,26,Oct-Dec 2023,Report AI calculation,,DERIVED,$1.7B / $6.5B x 100 = 26. Not used as the Jevons base year because Google's volume series has no 2023 value; the comparison stays at 2024 = 100. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,jevons_revenue_index,"Jevons check: revenue index, OpenAI + Anthropic run-rate revenue, 2024 = 100","index, 2024 = 100",2024,100,100,Dec 2024,Report AI calculation from press-reported run rates,,DERIVED,"Base: about $6.5B combined run-rate revenue = OpenAI about $5.5B (CNBC, Jun 2025, citing Dec 2024) + Anthropic about $1B (press reports). Run rate = latest month x 12; includes consumer subscriptions as well as token sales. Replaces the earlier Menlo-based spend index, whose 2026 value uses a new, non-comparable definition." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,jevons_revenue_index,"Jevons check: revenue index, OpenAI + Anthropic run-rate revenue, 2024 = 100","index, 2024 = 100",2025,446,446,end 2025,Report AI calculation,,DERIVED,"About $29B (OpenAI about $20B, page 935; Anthropic about $9B, Bloomberg) / $6.5B x 100 = 446. Price index on the same base (GPT-4-bar blended price, metric gpt4_bar_tokens_per_dollar): 66. Volume index: 4,948. Implied revenue per token = revenue index / volume index x 100 = 446 / 4,948 x 100 = 9.0." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,jevons_revenue_index,"Jevons check: revenue index, OpenAI + Anthropic run-rate revenue, 2024 = 100","index, 2024 = 100",2026,"1,769",1769,Jul-Sep 2026 (to date),Report AI calculation,https://thenextweb.com/news/anthropic-revenue-run-rate-65-billion,DERIVED,"About $115B (OpenAI about $50B end Sep 2026, CNBC; Anthropic more than $65B end Jul 2026, Bloomberg) / $6.5B x 100 = 1,769 (about 18x). Price index: 42; volume index: 32,990; implied revenue per token = 1,769 / 32,990 x 100 = 5.4, i.e. about 95% less revenue per token than in 2024. Caveats: two labs only; run rates are unaudited and dated two months apart; revenue includes subscriptions, while volume is Google's tokens, so the two legs do not cover the same activity. Read as an order of magnitude." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,revenue_per_million_tokens,"Revenue (spend) per million tokens routed, OpenRouter",US$ per 1M tokens,2022,,,,,,,Inputs not available. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,revenue_per_million_tokens,"Revenue (spend) per million tokens routed, OpenRouter",US$ per 1M tokens,2023,,,,,,,Inputs not available. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,revenue_per_million_tokens,"Revenue (spend) per million tokens routed, OpenRouter",US$ per 1M tokens,2024,,,,,,,Inputs not available. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,revenue_per_million_tokens,"Revenue (spend) per million tokens routed, OpenRouter",US$ per 1M tokens,2025,~$1.00,1.0,mid-2025,"Sacra (OpenRouter GMV about $100M annualised, May 2025) and a16z (more than 100T tokens per year, mid-2025); Report AI calculation",https://sacra.com/research/openrouter-100m-gmv/,DERIVED,"$100M / 100T tokens = $1.00 per 1M tokens of customer spend passing through OpenRouter to model providers. Second provider, ByteDance Volcano Engine 2025: MaaS revenue about RMB 1.5B (kr-asia) / about 8.6 quadrillion tokens (Report AI estimate from the 4T, 16.4T and 50T daily readings, linear between them) = about RMB 0.17, or about $0.024, per 1M tokens. That understates external revenue per token because Doubao's count includes internal and app traffic." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,revenue_per_million_tokens,"Revenue (spend) per million tokens routed, OpenRouter",US$ per 1M tokens,2026,~$0.71,0.71,Aug 2026 (to date),"Sacra (OpenRouter revenue about $160M annualised, Aug 2026; take rate about 5%) and Menlo Ventures (4.5 quadrillion tokens per year run rate, 19 Aug 2026); Report AI calculation",https://sacra.com/research/openrouter,DERIVED,"$160M / 0.05 = about $3.2B GMV; $3.2B / 4.5 quadrillion tokens = about $0.71 per 1M. Sensitive to the take-rate assumption (4% gives $0.89, 6% gives $0.59). Down about 29% from mid-2025, against OpenRouter's own statement that effective price per token fell 57-70% between January and October 2026. Same order as Silicon Data's $1.16-1.18 in early August 2026." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,average_request_cost_flagship,Cost of an average API request at OpenAI's flagship list price,US$ per request,2022,,,,,,,No published prompt-length data. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,average_request_cost_flagship,Cost of an average API request at OpenAI's flagship list price,US$ per request,2023,,,,,,,No published prompt-length data (the a16z / OpenRouter study starts Nov 2024). https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,average_request_cost_flagship,Cost of an average API request at OpenAI's flagship list price,US$ per request,2024,$0.0053,0.00525,Nov 2024,"a16z / OpenRouter 100 Trillion Token Study (average prompt about 1,500 tokens, completion about 150) at GPT-4o list price; Report AI calculation",https://a16z.com/state-of-ai/,DERIVED,"1,500 x $2.50/1M + 150 x $10/1M = $0.00375 + $0.0015 = $0.00525." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,average_request_cost_flagship,Cost of an average API request at OpenAI's flagship list price,US$ per request,2025,$0.0115,0.0115,Nov 2025,"Same study (average prompt about 6,000 tokens, completion about 400) at GPT-5 list price; Report AI calculation",https://a16z.com/state-of-ai/,DERIVED,"6,000 x $1.25/1M + 400 x $10/1M = $0.0075 + $0.004 = $0.0115, up 119% while the flagship's blended token price fell 21%. Prompt lengths are OpenRouter averages, not OpenAI's. Human comparison: OpenAI's GDPval (Sep 2025) found frontier models completed expert tasks about 100x faster and 100x cheaper than industry experts, counting inference time and API cost only, not review or rework." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,average_request_cost_flagship,Cost of an average API request at OpenAI's flagship list price,US$ per request,2026,,,,,,,"No 2026 average prompt length published. At 2025 lengths and GPT-6.1 Sol's $2 / $10, the same request would cost $0.016 (sensitivity, not a measurement)." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,roi_positive_wharton,Enterprise leaders reporting positive ROI from generative AI (Wharton / GBK),% of respondents,2022,,,,,,,Survey not run. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,roi_positive_wharton,Enterprise leaders reporting positive ROI from generative AI (Wharton / GBK),% of respondents,2023,,,,,,,No comparable positive-ROI reading found for Wharton's first (2023) wave. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,roi_positive_wharton,Enterprise leaders reporting positive ROI from generative AI (Wharton / GBK),% of respondents,2024,~65%,65,2024 survey,Wharton Human-AI Research / GBK Collective; Report AI calculation from the 2025 report's '+9 pp' change,https://knowledge.wharton.upenn.edu/special-report/2025-ai-adoption-report,DERIVED,74% - 9 pp = 65%. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,roi_positive_wharton,Enterprise leaders reporting positive ROI from generative AI (Wharton / GBK),% of respondents,2025,74%,74,Oct 2025,"Wharton / GBK, Accountable Acceleration (2025 AI Adoption Report), 800+ US enterprise decision-makers",https://knowledge.wharton.upenn.edu/special-report/2025-ai-adoption-report,MEDIUM,72% track structured ROI metrics. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,roi_positive_wharton,Enterprise leaders reporting positive ROI from generative AI (Wharton / GBK),% of respondents,2026,80% (28% strongly positive),80,13 Oct 2026 (to date),"Wharton / GBK, 2026 AI Adoption Report, as reported by HPCwire, Unite.AI and Beri",https://www.hpcwire.com/aiwire/2026/10/15/wharton-survey-finds-gaps-between-leaders-and-employees-on-ai-training-and-adoption/,MEDIUM,"Self-reported, not measured against spend. CONTESTED against MIT NANDA and IBM (see roi_shortfall_evidence)." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,roi_ebit_mckinsey,Organisations reporting any EBIT contribution from AI (McKinsey State of AI),% of respondents,2022,,,,,,,No comparable figure found. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,roi_ebit_mckinsey,Organisations reporting any EBIT contribution from AI (McKinsey State of AI),% of respondents,2023,,,,,,,No comparable figure found. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,roi_ebit_mckinsey,Organisations reporting any EBIT contribution from AI (McKinsey State of AI),% of respondents,2024,,,,,,,No comparable 2024 figure found. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,roi_ebit_mckinsey,Organisations reporting any EBIT contribution from AI (McKinsey State of AI),% of respondents,2025,39%,39,2025 survey (published Nov 2025),"McKinsey, The State of AI 2025",https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai,MEDIUM,As carried on page 393. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,roi_ebit_mckinsey,Organisations reporting any EBIT contribution from AI (McKinsey State of AI),% of respondents,2026,37%,37,survey 4 May - 8 Jun 2026,"McKinsey, The state of AI in 2026: On the road to ROI (1,719 respondents, 97 countries)",https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai,MEDIUM,"McKinsey calls it 'essentially unchanged from a year ago, at 37 percent', which suggests a restated 2025 base; our 39% for 2025 may be RESTATED. High performers (5%+ of EBIT from AI) about 6% in both years. About one in five organisations say AI operating costs now constrain use." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,roi_shortfall_evidence,Evidence of AI investments failing to return (share meeting expectations),% (definitions differ; see note),2022,,,,,,,No studies. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,roi_shortfall_evidence,Evidence of AI investments failing to return (share meeting expectations),% (definitions differ; see note),2023,,,,,,,No comparable study found. https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,roi_shortfall_evidence,Evidence of AI investments failing to return (share meeting expectations),% (definitions differ; see note),2024,,,,,,, https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,roi_shortfall_evidence,Evidence of AI investments failing to return (share meeting expectations),% (definitions differ; see note),2025,5% of pilots with measurable P&L impact (MIT NANDA); 25% of AI initiatives met expected ROI (IBM),5,Aug 2025 (MIT); May 2025 (IBM),"MIT Project NANDA, The GenAI Divide (Aug 2025); IBM Institute for Business Value CEO Study (May 2025, 2,000 CEOs)",https://newsroom.ibm.com/2025-05-06-ibm-study-ceos-double-down-on-ai-while-navigating-enterprise-hurdles,CONTESTED,"MIT defined success as measurable P&L movement within about six months (52 interviews, 153 surveys, 300 deployments); IBM measured initiatives against their own expected ROI. Wharton (74%) and Snowflake/ESG ($1.41 back per $1, Apr 2025) report the opposite picture. Different questions, samples and bars." https://report-ai.org/indexes/state-of-ai/state-of-ai-tokens/,2026-11-11,tokens,roi_shortfall_evidence,Evidence of AI investments failing to return (share meeting expectations),% (definitions differ; see note),2026,29% report significant ROI (Writer); 5-10% report full payback (PYMNTS),29,2026,"Writer, Enterprise AI adoption in 2026; PYMNTS Intelligence (Jun 2026 survey of 60 executives)",https://writer.com/blog/enterprise-ai-adoption-2026/,CONTESTED,"Snowflake/Omdia's 2026 survey (2,050 respondents) reports average ROI of 49%, or $1.49 per $1 spent, up from $1.41 in 2025; Wharton 2026 reports 80% positive but 28% strongly positive. No study measures return per token."