“while prices for some models might have come down, newer “reasoning” models burn way more tokens, which means the cost of inference has, somehow, gotten higher over time.”
Our reading, the sentence that was checked:
Because reasoning models consume far more tokens, the cost of inference has risen over time even if some per-model prices fell.
Assessment: Unresolved Evidence as of
Where and when Ed Zitron wrote it
Where's Your Ed At, AI's Economics Don't Make Sense [Ad Free], paragraph 20.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment wyea-ais-economics-dont-make-sense-ad-free-b79382-a1
We recorded 28 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2026-04-28. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Causal claim
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No horizon
- Deadline state on 2026-10-29
- No horizon (the release’s reference date)
Conditions
- reasoning models burn more tokens (qualifier, stated by the speaker)
From the speaker
- prices for some models might have come down
- newer “reasoning” models burn way more tokens
- which means the cost of inference has, somehow, gotten higher over time
Our interpretation
- The ordinary task-level reading is that additional reasoning-token consumption outweighed price declines and raised the cost of completing comparable work as newer models were adopted.
- A lower per-token price alone does not contradict this claim.
- Higher expenditure on harder tasks, more tasks, or higher-quality answers does not by itself establish a rise for comparable work.
- The speaker supplies no starting date. The assessment covers the historical trend through publication without inventing a starting boundary.
- This is a general inference-cost claim rather than a claim about one company's accounts, so metric_scope is null.
Counts as supported if: Comparable longitudinal task measurements show increased total inference costs, with token counts and prices demonstrating that increased reasoning caused the rise.
Counts as contradicted if: Representative measurements of comparable work show that total inference costs declined despite reasoning-model adoption, after accounting for reasoning tokens and relevant prices.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
Outcome: Unresolved
The record was checked, and the evidence available by the cutoff does not settle it.
Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.
- Evidence as of
- Deadline state when assessed
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The proposed mechanism is credible and documented: longer reasoning can raise the bill for a request. Research also finds rising expenditure on frontier models. But the same research finds declining costs at fixed performance and roughly constant average costs on one coding benchmark. These are different comparisons, not evidence that the entire claim is both true and false. The broad historical trend remains unsettled without a defined workload and cost measure.
Limitations
- The main empirical source is a preprint.
- Benchmark workloads may not represent production subscriptions.
- Choosing frontier capability rather than fixed performance can change the result.
Evidence (2 items)
Context only: Kilo,
test-time scaling
Why it matters here: Kilo describes additional reasoning and longer contexts as sources of higher application expenditure. Its discussion does not hold task scope, output quality, or usage constant, so it does not establish the general comparable-task trend.
- Publisher
- Kilo
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: arXiv; Hans Gundlach, Jayson Lynch, Matthias Mertens, and Neil Thompson,
For SWE-bench Verified, average evaluation prices have remained roughly constant.
Why it matters here: Using price data through November 2025, the paper finds rising frontier expenditure but falling costs at fixed performance, while average SWE-bench evaluation prices remain roughly flat. These results support a narrower frontier-cost reading and challenge a general increase, without measuring the actual workload mix of AI services.
- Publisher
- arXiv; Hans Gundlach, Jayson Lynch, Matthias Mertens, and Neil Thompson
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
A defined task basket and longitudinal measurements of total cost, token consumption, and reasoning settings, with consistent quality requirements and a stated price-versus-expense measure.
Where it was said (1)
Said once in the sources this register covers.
: Where's Your Ed At, AI's Economics Don't Make Sense [Ad Free]
while prices for some models might have come down, newer “reasoning” models burn way more tokens, which means the cost of inference has, somehow, gotten higher over time.
- Where
- paragraph 20
- Published
- 2026-04-28
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment wyea-ais-economics-dont-make-sense-ad-free-b79382-a1: Current. Outcome: Unresolved. Evidence as of 2026-10-28. Public date not recorded.
Recorded events
Assessment publishedBy this site
First assessment published after both reviews and the final-text check passed.
About assessment wyea-ais-economics-dont-make-sense-ad-free-b79382-a1.