“It is still expensive to operate ("inference"), especially if you want to make it "think" longer like o3.”
Our reading, the sentence that was checked:
DeepSeek r1 is still expensive to run at inference, especially when made to reason longer like o3.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Five things most people don't seem to understand about DeepSeek, paragraph 3.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-five-things-most-people-dont-seem-507f10-a1
We recorded 9 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 2025-01-28. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Observation
- 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
- especially when made to think longer like o3 (qualifier, stated by the speaker)
From the speaker
- still expensive to operate
- inference
- especially if you want to make it "think" longer like o3
Our interpretation
- The ordinary reading concerns the material computing expense of serving the full original R1 model, especially per completed answer when reasoning is extended.
- The reference to o3 illustrates extended reasoning; it does not assert that R1 costs more than o3.
- Longer thinking is an emphasized operating scenario, not a conditional forecast.
- Use the release-to-post period for this ongoing observation. No workload, cost threshold or accounting basis is supplied.
Counts as supported if: Matched measurements establish substantial operating expense for ordinary R1 workloads and greater expense per completed answer when reasoning is extended.
Counts as contradicted if: Matched measurements establish low operating expense under an appropriate ordinary workload and refute the claimed cost increase from extended reasoning.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
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 engineering evidence supports the mechanism that longer reasoning consumes more resources. It does not settle whether January R1 operation was expensive under a specified workload. The direct cost disclosure concerns February and combines V3 with R1; lower API prices measure a different quantity. An unknown accounting basis also prevents a supported or contradicted money assessment.
Limitations
- The claim has a meaningful ordinary reading, but its cost level remains unmeasured.
- A large deployment's daily expense alone does not establish high cost per useful answer.
- Later serving optimizations cannot automatically be attributed to the January release period.
Evidence (3 items)
Context only: NVIDIA,
The DeepSeek-R1 NIM microservice can deliver up to 3,872 tokens per second on a single NVIDIA HGX H200 system.
Why it matters here: NVIDIA describes an eight-H200 deployment and explains that extended reasoning generates more tokens and longer generation cycles. This supports the resource mechanism, but neither measures monetary expense nor falls inside the target period.
- Publisher
- NVIDIA
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: DeepSeek-AI,
Assuming the leasing cost of one H800 GPU is $2 per hour, the total daily cost amounts to $87,072.
Why it matters here: This is a rental-equivalent daily compute estimate for V3 and R1 together. It cannot establish January R1-only operating expense or the incremental cost of longer reasoning.
- Publisher
- DeepSeek-AI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Measures
- DeepSeek, Combined V3 and R1 inference GPU cost at assumed rental rates, USD dollars per day, company reported, 2025-02-27 to 2025-02-28
- Describes events of
- Public from
Context only: Epoch AI,
However, while the performance difference between the two models is small, the pricing difference is definitely not.
Why it matters here: The analysis reports much lower customer output-token pricing for R1 than o1. This cautions against interpreting expensive as a relative API-price claim, but retail prices do not disclose provider costs.
- Publisher
- Epoch AI
- Kind of source
- Secondary summary
- Measures
- DeepSeek, Customer API output-token price for R1, USD dollars per million output tokens, estimate, 2025-01-31 to 2025-01-31
- Describes events of
- Public from
What would settle it
January R1-only serving-cost measurements for defined workloads, including matched short- and long-reasoning runs, together with a specified cost basis and comparison for expensive.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Five things most people don't seem to understand about DeepSeek
It is still expensive to operate ("inference"), especially if you want to make it "think" longer like o3.
- Where
- paragraph 3
- Published
- 2025-01-28
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-five-things-most-people-dont-seem-507f10-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 marcus-five-things-most-people-dont-seem-507f10-a1.