Toby Ord · Dec 22, 2025
Toby Ord argues AI agents’ costs may outpace their capabilities
Markdown · JSON · Original source
Toby Ord’s article was published on December 22, 2025. Ord argues that the trend does not show whether the tasks can perform are economical. He analyzes METR charts by identifying each model’s “sweet spot,” where its hourly cost is cheapest, and an estimated “saturation point,” where additional compute yields sharply diminished gains. In his analysis, models with longer time horizons tend to have higher hourly costs. Some cost as much as or more than human labor near their full time horizon. Ord calls this “moderate evidence” that costs, including hourly costs, are rising exponentially. If the relationship continues, he argues, forecasts based on time horizons alone could overstate when AI agents will become economically feasible. Practical use may lag behind measured capabilities. Ord cautions that model price data may be too high. He says the analysis needs more representative models, better treatment of chart upticks, and explicit cost-over- analysis. Ord concludes that the gap between what AI agents can achieve in principle and what is economically feasible may grow. How hourly costs change over time remains the key question.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
Toby Ord · Article
Source details
Original source date: Dec 22, 2025 (2025-12-22T18:30:04+00:00).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-11-09T13:08:25.026Z. First observed: 2026-10-25T19:27:58.001Z. Retrieved: 2026-10-25T19:29:48.736Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
Text as retrieved on October 25, 2026; originally published December 22, 2025. Ord describes the analysis as having significant limitations: it uses a limited and potentially unrepresentative set of models, has unresolved chart upticks and possible cost-data issues, and does not explicitly plot hourly cost against release date or numerically determine trends and correlations. The source says its saturation-point definition is somewhat arbitrary and that the final chart upticks complicate analysis. Ord says the price data for OpenAI models was still an estimate, which could make their plotted costs too high.
Open question
Will a broader analysis of model costs over show that ’ hourly costs rise with their measured ?
Source-stated
Ord argues that METR’s time-horizon results do not by themselves reveal the cost of achieving the measured performance.
Their headline time-horizon numbers are meant to show the best possible performance that can be attained with a model (regardless of cost).
Source-stated
Ord reports a positive correlation between task duration and hourly cost in the analyzed models, both at sweet spots and at saturation points.
Again, there is a correlation between time horizon and cost, and again the hourly costs seem to be increasing with time horizon too.
Source-stated
Ord calls the analysis moderate evidence that hourly costs are rising exponentially, while warning that key parts of the analysis need further work.
the costs to achieve the time horizons are growing exponentially, even the hourly costs are rising exponentially, the hourly costs for some models are now close to human costs.
Source-stated
Ord argues that if the cost relationship continues, forecasts based on METR’s time-horizon trend may overstate when capabilities will be practical.
If this relationship were to continue, then forecasting when certain time horizons will be available from the headline METR trend will be misleading, as the models would be impractically expensive when they first reach those capabilities.
Source-stated
Ord says OpenAI model price estimates may be too high, so their apparent costs relative to human labor should not be overemphasized.
The METR page suggests that the price data for those models was still an estimate at that point (based on o1 costs), so I wouldn’t be surprised if these curves should really be shifted somewhat to the left, making them several times cheaper.
Source and contributor context
Toby Ord · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
AI agent
An AI system that chooses and carries out actions toward a goal, often using software tools.
Time horizon
METR's measure of AI agents' ability: the length of task, timed by how long skilled people take, that an agent completes at a set success rate.
METR
METR studies what AI systems can do and how reliably they complete tasks, including work with human developers.
OpenAI
OpenAI develops AI models and products and publishes research, release announcements and evaluations.
TIME
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Editorial provenance
Version 1. Publication ID: c48421de-e32e-2ea1-5870-1b54af06502c.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-09T13:03:47.365Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: bffdbafa035700177cef174484939b775049e4b29bfc0502209c0628b0a5932a. Source text hash: fbb9efe0de68539d3bd6673fc0af61ae64219287258da01b74ba0afe730c8e9f.
Revision and source-check history
Version 1 · 2026-11-09T13:08:25.026Z
Toby Ord argues AI agents’ costs may outpace their capabilities
Toby Ord’s article was published on December 22, 2025. Ord argues that the trend does not show whether the tasks can perform are economical. He analyzes METR charts by identifying each model’s “sweet spot,” where its hourly cost is cheapest, and an estimated “saturation point,” where additional compute yields sharply diminished gains. In his analysis, models with longer time horizons tend to have higher hourly costs. Some cost as much as or more than human labor near their full time horizon. Ord calls this “moderate evidence” that costs, including hourly costs, are rising exponentially. If the relationship continues, he argues, forecasts based on time horizons alone could overstate when AI agents will become economically feasible. Practical use may lag behind measured capabilities. Ord cautions that model price data may be too high. He says the analysis needs more representative models, better treatment of chart upticks, and explicit cost-over- analysis. Ord concludes that the gap between what AI agents can achieve in principle and what is economically feasible may grow. How hourly costs change over time remains the key question.
Original source
Toby Ord · Article
Source details
Original source date: Dec 22, 2025 (2025-12-22T18:30:04+00:00).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-11-09T13:08:25.026Z. First observed: 2026-10-25T19:27:58.001Z. Retrieved: 2026-10-25T19:29:48.736Z.
Limits
Text as retrieved on October 25, 2026; originally published December 22, 2025. Ord describes the analysis as having significant limitations: it uses a limited and potentially unrepresentative set of models, has unresolved chart upticks and possible cost-data issues, and does not explicitly plot hourly cost against release date or numerically determine trends and correlations. The source says its saturation-point definition is somewhat arbitrary and that the final chart upticks complicate analysis. Ord says the price data for OpenAI models was still an estimate, which could make their plotted costs too high.
Open question
Will a broader analysis of model costs over show that ’ hourly costs rise with their measured ?
Source-stated
Ord argues that METR’s time-horizon results do not by themselves reveal the cost of achieving the measured performance.
Their headline time-horizon numbers are meant to show the best possible performance that can be attained with a model (regardless of cost).
Source-stated
Ord reports a positive correlation between task duration and hourly cost in the analyzed models, both at sweet spots and at saturation points.
Again, there is a correlation between time horizon and cost, and again the hourly costs seem to be increasing with time horizon too.
Source-stated
Ord calls the analysis moderate evidence that hourly costs are rising exponentially, while warning that key parts of the analysis need further work.
the costs to achieve the time horizons are growing exponentially, even the hourly costs are rising exponentially, the hourly costs for some models are now close to human costs.
Source-stated
Ord argues that if the cost relationship continues, forecasts based on METR’s time-horizon trend may overstate when capabilities will be practical.
If this relationship were to continue, then forecasting when certain time horizons will be available from the headline METR trend will be misleading, as the models would be impractically expensive when they first reach those capabilities.
Source-stated
Ord says OpenAI model price estimates may be too high, so their apparent costs relative to human labor should not be overemphasized.
The METR page suggests that the price data for those models was still an estimate at that point (based on o1 costs), so I wouldn’t be surprised if these curves should really be shifted somewhat to the left, making them several times cheaper.
Source and contributor context
Toby Ord · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
AI agent
An AI system that chooses and carries out actions toward a goal, often using software tools.
Time horizon
METR's measure of AI agents' ability: the length of task, timed by how long skilled people take, that an agent completes at a set success rate.
METR
METR studies what AI systems can do and how reliably they complete tasks, including work with human developers.
OpenAI
OpenAI develops AI models and products and publishes research, release announcements and evaluations.
TIME
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Editorial provenance
Version 1. Publication ID: c48421de-e32e-2ea1-5870-1b54af06502c.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-09T13:03:47.365Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: bffdbafa035700177cef174484939b775049e4b29bfc0502209c0628b0a5932a. Source text hash: fbb9efe0de68539d3bd6673fc0af61ae64219287258da01b74ba0afe730c8e9f.
uncertain · 2026-11-09T13:03:47.365Z · anthropic/claude-opus-5.5
Publication: c48421de-e32e-2ea1-5870-1b54af06502c
attribution — pass: The page is a single essay by Toby Ord on his own site, not a listing. The reading attributes the argument, the analysis of METR's charts and the conclusions to Ord, and keeps METR's role as the source of the underlying charts and data distinct.
Are the Costs of AI Agents Also Rising Exponentially? December 22, 2025Toby Ord
summary — uncertain: The summary is faithful and keeps Ord's hedges, such as 'moderate evidence' and 'if the relationship continues'. However, METR, the organization whose trend and charts the whole argument rests on, is never identified. A reader without background is not told it is the group that measures AI agents' task lengths. 'Chart upticks' is also used without explanation. Smallest repair: add a brief phrase saying what METR is, and gloss upticks as late jumps caused by models withholding answers until the last moment.
We’ve all seen the graphs from METR showing that the length of tasks AI agents can perform has been growing exponentially over the last 7 years.
date — pass: The reading's date, December 22, 2025, matches the UTC day of originalPublished and the source's own dateline. Nothing is presented as known after that date, and the forecasts remain Ord's conditional claims. The page was modified on 2026-01-16, but the reading relies on core analysis that is plainly part of the original piece, not on an apparent later addition. This is a publisher page, so no archive disclosure is required.
December 22, 2025Toby Ord
support — pass: Each claim matches the source: the correlation of hourly cost with time horizon, 'moderate evidence' of exponential cost growth, the conditional forecasting caveat, the caveat on OpenAI pricing, and the listed limitations. The phrase 'Some cost as much as or more than human labor' is supported by o3's $350 per hour. The page was edited after publication, but the cited passages are plainly part of the main essay.
And o3 actually costs \$350 per hour (more than the human price) to achieve tasks at its full 1.5 hour task horizon.
media — pass: No media is selected. No verified candidate media were supplied, so no mismatch is possible.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
METR studies what AI systems can do and how reliably they complete tasks, including work with human developers.
Explore METRMETR's measure of AI agents' ability: the length of task, timed by how long skilled people take, that an agent completes at a set success rate.
Definition referenceAn AI system that chooses and carries out actions toward a goal, often using software tools.
Definition referenceOpenAI develops AI models and products and publishes research, release announcements and evaluations.
Explore OpenAITIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEOpenAI develops AI models and products and publishes research, release announcements and evaluations.
Explore OpenAIMETR studies what AI systems can do and how reliably they complete tasks, including work with human developers.
Explore METRTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEAn AI system that chooses and carries out actions toward a goal, often using software tools.
Definition referenceMETR's measure of AI agents' ability: the length of task, timed by how long skilled people take, that an agent completes at a set success rate.
Definition referenceMETR studies what AI systems can do and how reliably they complete tasks, including work with human developers.
Explore METRMETR's measure of AI agents' ability: the length of task, timed by how long skilled people take, that an agent completes at a set success rate.
Definition referenceAn AI system that chooses and carries out actions toward a goal, often using software tools.
Definition referenceOpenAI develops AI models and products and publishes research, release announcements and evaluations.
Explore OpenAITIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEOpenAI develops AI models and products and publishes research, release announcements and evaluations.
Explore OpenAIMETR studies what AI systems can do and how reliably they complete tasks, including work with human developers.
Explore METRTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEAn AI system that chooses and carries out actions toward a goal, often using software tools.
Definition referenceMETR's measure of AI agents' ability: the length of task, timed by how long skilled people take, that an agent completes at a set success rate.
Definition reference