Urdais

Model Economics

The economics of machine intelligence.

Price, consumption, market share, and capability across the model economy.

Token Price

Cost of a standardized 1M-token workload

$30.00 per 1M tokens

    Observed Token Volume Index

    Observed model consumption over time

    UTVI · Observed Token Volume Index · updated Oct 29, 2026 · 00:00 UTC · Provisional

    33.10T tokens/day +12.4% 1D

    Token volume exposed by OpenRouter's rankings-daily dataset for the traffic included by that dataset. Urdais makes no claim about inclusion of BYOK or hidden/private application traffic unless OpenRouter explicitly documents it.

      Source: OpenRouter, as of 2026-10-30T02:58:28.000Z. Licensed under CC BY 4.0. Methodology UTVI 1.1.0. The source published no data for 2025-06-15 and 2025-07-15; those dates carry no value and none is interpolated.

      Market Share

      Share of observed OpenRouter token volume represented in UTVI.

      2026-10-29 · denominator 33.10T observed tokens · Provisional

      1. 01DeepSeek34.51%11.42T
      2. 02OpenAI14.44%4.78T
      3. 03Z.ai (Zhipu AI)10.07%3.33T
      4. 04Anthropic7.52%2.49T
      5. 05Tencent6.46%2.14T
      6. 06StepFun3.78%1.25T
      7. 07MiniMax3.09%1.02T
      8. 08NVIDIA2.84%940.36B
      9. 09Google2.29%757.22B
      10. 10Xiaomi1.82%603.32B
      11. 11Moonshot AI1.82%602.13B
      12. 12Meta1.19%394.90B
      13. 13Alibaba Cloud0.79%262.90B
      14. 14Upstage0.58%190.38B
      15. 15Poolside0.37%123.31B
      16. 16Perplexity0.17%57.83B
      17. 17Thinking Machines0.17%55.26B
      18. —Unattributed models3 named models with no evidenced lab2.38%786.47B
      19. —Other modelsOpenRouter residual: volume outside the models the dataset names5.69%1.88T

      Token volume exposed by OpenRouter's rankings-daily dataset for the traffic included by that dataset. Urdais makes no claim about inclusion of BYOK or hidden/private application traffic unless OpenRouter explicitly documents it.

      Source: OpenRouter rankings-daily, as of 2026-10-30T02:58:28.000Z. Licensed under CC BY 4.0. Methodology Market Share, under UTVI 1.1.0. Shares are observational: the denominator is the traffic this one dataset exposes, not global model usage. Named models whose lab Urdais cannot evidence are reported as unattributed rather than assigned, and OpenRouter’s own residual row is never credited to a lab.

      Model Frontier

      Benchmark capability against provider list price per 1M tokens.

      45 configurations across 25 models · 3 on the frontier · prices as of 2026-09-22 · scores evaluated 2025-10-16 to 2026-09-22

      Model Frontier compares benchmark capability with provider list price per 1M tokens. Reasoning effort may change the number of tokens consumed, so equal unit token prices do not imply equal total cost per request or task.

      Not plotted on GPQA Diamond: 280 scored models Urdais has not linked to a priced product, 0 whose identity is ambiguous, and 0 linked models with no comparable standard list price. They are excluded rather than estimated.

      Capability: Epoch AI, 'Capabilities & Benchmarking'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/benchmarks' [online resource]. Licensed under CC BY 4.0. Price: Urdais Token Price, a 500k input plus 500k output workload weighted evenly. Methodology Model Frontier 1.0.0. A point is one model under one source-declared configuration; capability and price carry different dates and are not reconciled.

      Open-weight vs Proprietary

      Whether each model's publisher released downloadable weights, measured across observed OpenRouter token volume, Epoch AI benchmark scores, and published list prices.

      Volume share

      Observed token volume, 30 days to 2026-10-29

      Open-weight(14 models)
      42.6%
      Proprietary(5 models)
      2.5%
      Unclassified
      54.9%

      Unclassified is 48.9T in the source’s own residual row, which names no model; 0.0T under aliases and anonymous endpoints that cannot be resolved to the model that served them; 363.9T under identifiers Urdais has not mapped yet; 2.5T whose model is known but whose access is not; and 0.0T whose weights are published for non-commercial use only. It is reported rather than redistributed.

      Capability gap

      Best-scoring configuration in each class, GPQA Diamond

      Proprietary Gemini 3.7 Flash
      94.8%
      Open-weight Kimi K3
      93.1%
      Gap
      1.7%proprietary leads

      Price gap

      Median blended list price among Pareto-efficient configurations, GPQA Diamond

      Proprietary (n = 1)
      $0.70
      Open-weight (n = 0)
      —
      Ratio
      Insufficient comparable frontier coverage

      Highest-scoring configuration in each class: Gemini 3.7 Flash at $2.25, Kimi K3 at $9.00 per 1M tokens.

      Prices as of 2026-09-22

      Open-weight means the publisher offers the weights for download under a licence permitting commercial use; it does not mean open source, and several of these licences restrict it. Weights published for non-commercial use only are reported as Unclassified. Volume is share of observed OpenRouter token volume, not of the industry. Prices are published list prices per token, not total inference cost: a model that emits more reasoning tokens to answer the same question costs more than its per-token price suggests.