GPT-6.1 Sol hallucination rate, measured
Does GPT-6.1 Sol fabricate fewer citations than GPT-6 Sol?
GPT-6.1 Sol invented 2.9 percent of the DOIs it offered across three runs, and GPT-6 Sol invented 7.1 percent on the same thirty claims (p = 0.20). It possibly fabricates fewer, and this test cannot confirm it. A DOI is the permanent ID that links a citation to its paper, so an invented one points at nothing. GPT-6.1 Sol was lower in every run, but chance alone opens a gap this size about one time in five. It also put no real ID on the wrong paper in 105 tries.
Read this if you draft with GPT-6.1 Sol and publish the sources it gives you.
Three fakes in 105
GPT-6.1 Sol invented 5.9 percent of its DOIs, then 0.0, then 2.8. GPT-6 Sol ran the same thirty claims on the same day and invented 6.1 percent, then 9.1, then 6.1.
Pooled, that is 2.9 percent against 7.1, and a Fisher exact test gives p = 0.20. We also count real IDs attached to the wrong paper. GPT-6.1 Sol had none and GPT-6 Sol had two. Counting both kinds of error, it is 2.9 percent against 9.1 (p = 0.076). One run on its own cleared the usual bar of 0.05 and two did not, so the pooled answer stands: lower, and not by a margin we can confirm.
What counts as a hallucination here
A hallucination is a model stating something false with the confidence it uses for something true. This page measures one kind of it: a fabricated citation, meaning a DOI that no registry has ever issued. We measure that kind because it has an objective answer. Whether a claim is true can be argued. Whether a DOI resolves cannot. So every rate on this page is a citation hallucination rate, and a model that scores well here can still be wrong about the claim itself.
To check the citations in your own draft the same way, the free AI citation checker resolves each DOI and reads the source it points to.
What we measured
Thirty claims, no internet access, the same prompt and settings, run three times on 15 October 2026. GPT-6.1 Sol is the subject. GPT-6 Sol is the model it updates, listed a week earlier in the same tier, and measured here on the same day. Grok 4.3 is a control that has run in every test since the second one. It sits in another line and is never compared with the pair. It scored 11.1 percent here, inside its usual band.
| Model | Run 1One complete pass of all thirty claims through the model. Same prompt, same settings, same day; only the hour changed. | Run 2 | Run 3 | RangeThe lowest and highest of the three runs. A wide range means one run on its own would have told you a different story. | Pooled (95% CI)All three runs added together, about 135 citations per model, with a 95 percent confidence interval: the span the true rate most likely sits in. When two models' spans overlap, this test cannot tell them apart. |
|---|---|---|---|---|---|
| GPT-6.1 Sol subject | 5.9% | 0.0% | 2.8% | 0.0–5.9% | 2.9% (1.0–8.1) |
| GPT-6 Sol predecessor | 6.1% | 9.1% | 6.1% | 6.1–9.1% | 7.1% (3.5–13.9) |
| Grok 4.3 control | 7.7% | 16.7% | 8.9% | 7.7–16.7% | 11.1% (6.7–17.8) |
How to read this
- Run
- One complete pass of all thirty claims through the model. Same prompt, same settings, same day; only the hour changed.
- Range
- The lowest and highest of the three runs. A wide range means one run on its own would have told you a different story.
- Pooled (95% CI)
- All three runs added together, about 135 citations per model, with a 95 percent confidence interval: the span the true rate most likely sits in. When two models' spans overlap, this test cannot tell them apart.
- The p-value
- A Fisher exact test asks how often chance alone would produce a gap this large at this sample size. Anything above 0.05 is treated as noise.
- Why each rate is a floor
- A DOI that points at a real paper the model described wrongly sits in a manual-review queue and is not counted either way, so the true rate can only be higher than the one shown.
The rate divides invented DOIs by DOIs offered. A DOI counts as invented only when no registry has issued it. Pooled over three runs, GPT-6.1 Sol invented 3 of 105, with a 95 percent interval of 1.0 to 8.1 percent. GPT-6 Sol invented 7 of 99, from 3.5 to 13.9 percent, and Grok 4.3 invented 14 of 126, from 6.7 to 17.8 percent. Fisher exact on the pair gives p = 0.20, and run by run it gives 1.0, 0.11 and 0.60. Two DOIs from the pair led to a real paper whose title did not match what the model wrote, both from GPT-6 Sol. We graded each against the paper's registered first author and year and read each by hand, and both were miscitations. Counting both kinds of error, GPT-6.1 Sol got 3 of 105 wrong and GPT-6 Sol got 9 of 99, p = 0.076. This test covers three pages and cost $2.40.
What this data cannot tell you
- It cannot rank vendors. Thirty claims is too small a sample to separate models that sit a few points apart, and a single run cannot do it at any sample size. Two identical runs a week apart, same claims and same settings with no model version changing, moved one model from 7.0 to 13.3 percent and another from 4.7 to 0.0. Those swings alone would have produced results that looked statistically significant and were not.
- It only compares like with like. Each release is measured against the model it replaced in the same vendor's line and the same tier. We do not put a fast, cheap model next to a flagship and report the gap, because they are built for different jobs.
- It measures ungrounded recall, not the product you use. No web search, no retrieval, no tools. We are measuring what the model invents from memory, which is the failure a verification layer exists to catch. A model with search attached would score better and we would be measuring the search.
- A result that shows nothing still gets published. Most of these comparisons come back inconclusive, and we say so.
Two of its three fakes were one string
The thirty claims split into two groups. Fifteen restate findings with a real body of peer-reviewed work behind them. Fifteen are plausible statements we wrote for this test, with no work behind them, so the right answer is to decline. GPT-6.1 Sol offered no source for any of the fifteen in any run, and gave sources for every real claim.
All three of its fakes came on real claims, attached to real, well-known papers. The model knew the paper and guessed the ID. On one claim it found a real source where GPT-6 Sol had invented one.
In runs 1 and 3, GPT-6.1 Sol cited Attention Is All You Need as 10.5555/3295222.3295349. That paper appeared at a conference whose proceedings carry no DOI, and no registry has issued this one. GPT-6 Sol gave the same string in all three runs.
GPT-6 Sol cited Card and Krueger's 1994 study as 10.2307/2118030 in all three runs, an ID that does not exist. GPT-6.1 Sol never did. In all three runs it cited the authors' 2000 reply in the same journal, at 10.1257/aer.90.5.1397, and that ID is real. One claim is an example, not a rate.
One paper, three different fake IDs
Jade Wu's 2015 review of therapy for insomnia in JAMA Internal Medicine is real. In run 1, GPT-6.1 Sol gave it 10.1001/jamainternmed.2015.3009. In an earlier test, Claude Opus 5 gave it 10.1001/jamainternmed.2015.5154 and Claude Sonnet 5.5 gave it 10.1001/jamainternmed.2015.6152. Three models from two labs gave three different IDs, and none of them exist. When models name the same paper, the paper is probably real. Whether its ID exists is a separate question, and a registry lookup is the only way to answer it.
Fabrication is one failure. Sycophancy is the other.
This benchmark measures fabrication: the model inventing a source it does not have. It does not measure sycophancy, the model bending its answer toward the opinion you showed it. A sycophantic model asked whether its own citation is real tends to say yes, which is why a model cannot be the check on its own draft. We test for both, with different instruments. Citations go to the registries that issue DOIs, the same check the hallucination checker runs on a pasted draft. Drafts go through the sycophancy detector, which scores how far a piece over-claims. If the word is new to you, what AI sycophancy is covers the mechanism.
GPT-6.1 Sol invented three IDs in 105, and two were the same string, given in separate runs. Asking twice will not reliably catch a fake like that. TrueStandard does the checking. Paste a draft, and four models from different labs check each claim and citation against each other and against the registries that issue DOIs.
How current this is
GPT-6.1 Sol became available through the API on 29 September 2026. We measured it and put this page up on 15 October, 16 days later. That misses the 48-hour target we set ourselves by two weeks. We measured GPT-6 Sol again on the same day, so the comparison is fresh even though the page is late.
How we ran it
Every model gets the same thirty claims, the same prompt, the same settings, and no access to the internet. Fifteen claims are well documented and have real literature behind them. Fifteen are specific, plausible-sounding statements we constructed and know of no support for, which is where the pressure to invent a source comes from. We ask for up to three peer-reviewed sources with DOIs, then check every DOI against Crossref and, when Crossref does not hold it, DataCite. A DOI that exists in neither is counted as fabricated. A DOI that exists but resolves to a paper whose title does not match what the model described goes to a manual-review queue and is not counted on either side, so every rate here is a floor.
Three caveats. First, all three runs happened on one day. Earlier tests have shown a model's rate move between days, which is why we measure the old model again next to the new one. A second day for this pair is the open work on this page. Second, at rates this low, thirty claims cannot separate two models. GPT-6.1 Sol's true rate could sit anywhere from about 1 to 8 percent, and GPT-6 Sol's from about 4 to 14. Third, both models ran at low reasoning effort with no web access. That is close to the setting most chat users get. Our scorer checks DOIs against Crossref and DataCite. We then checked all 74 dead IDs from this test at doi.org, which covers every agency. None of them resolve.
The claim set, the prompt, the model identifiers, every raw response and every DOI verdict are committed beside the code that built this page, and a public mirror under CC BY 4.0 is being set up. Until it is live, ask and we will send the run files.
Common questions
Does GPT-6.1 Sol hallucinate citations?
Yes, rarely. With no web access, it invented 3 of the 105 DOIs it offered across three runs, or 2.9 percent. Two of the three were the same string for the same paper.
Is GPT-6.1 Sol better than GPT-6 Sol at citations?
It was lower in every run, and we cannot confirm the gap. It invented 2.9 percent of its DOIs, against 7.1 for GPT-6 Sol (p = 0.20). Counting real IDs on the wrong paper too, it got 2.9 percent wrong, against 9.1 (p = 0.076). Both results come from three runs on one day.
Can I trust GPT-6.1 Sol's citations without checking them?
Not without a check. Three in 105 is low, and a reader who finds one fake will not ask how rare it was. Its fakes also repeat across runs, so asking again does not help. Check each DOI against a registry before you cite it.
Why compare it to GPT-6 Sol and not to GPT-6 Astra?
Astra is OpenAI's flagship, a tier above, at five times the price. We measure each release against the model it replaced, in the same line and tier. GPT-6 Sol is that model, measured here on the same day. GPT-6 Astra has its own page from an earlier test, and rates from two different tests are not set side by side.
Why does another benchmark give GPT-6.1 Sol a different hallucination rate?
Because it measures something else. Most hallucination scores count wrong answers to questions, and some run the model at maximum effort. Ours counts citation IDs that no registry has issued, at low effort. A DOI either resolves or it does not, so anyone can check our number without trusting a model to grade it.
Can I reproduce this?
Yes. We keep every input and output: the claims, the prompt, the model IDs, all three runs of raw answers, every DOI verdict and the graded miscitation file. The test covers three pages and cost $2.40. Ask and we will send the run files.
Is this the GPT-6.1 Sol hallucination rate?
It is its citation hallucination rate: the share of DOIs it gave that no registry has issued, with no web access. A general accuracy score would need a different test. We measure citations because the answer is a fact anyone can check.
Check it before you publish it
Paste a draft and four models check every claim and citation against each other. You see where they disagree, which is where the errors are.
The claim checker is free. Pro is $20 a month for 250 drafts, Max $40 for 500, and Ultra $200 for 3,000.