GLM-5.3 Prime hallucination rate, measured
Does GLM-5.3 Prime make up fewer citations than GLM-5.3?
GLM-5.3 Prime made up 29.6 percent of the DOIs it offered across three runs, and GLM-5.3 made up 31.5 percent on the same thirty claims (p = 0.79). Chance explains a gap that small. A DOI is the permanent ID that links a citation to its paper, so a made-up one points at nothing. Counting real IDs put on the wrong paper too, Prime got 47.4 percent wrong against 63.0 for GLM-5.3. Chance would open that gap about 13 times in 1,000 (p = 0.013). Prime is listed as a faster version of the same model, and we know of no reason it would make fewer miscitations. All three runs were on one day, so read it as a measured gap with no known cause until a second day repeats it.
Read this if you draft with GLM-5.3 Prime and publish the sources it gives you.
Level on made-up IDs, apart on miscitations
GLM-5.3 Prime made up 37.8 percent of its DOIs, then 26.7, then 24.4. GLM-5.3 ran the same thirty claims on the same day and made up 27.9 percent, then 27.9, then 41.5.
Pooled, that is 29.6 percent against 31.5, and a Fisher exact test gives p = 0.79. Prime was higher in run 1 and lower in runs 2 and 3. On made-up IDs, that is a null result. We also count real IDs attached to the wrong paper. Prime had 24 and GLM-5.3 had 39. Counting both kinds of error, it is 47.4 percent against 63.0, and p = 0.013. That clears the usual bar of 0.05. We report the gap and do not claim a cause.
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 23 October 2026. GLM-5.3 Prime is the subject. GLM-5.3 is the same model at its usual speed, and we measured it 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 8.7 percent here, inside its usual band. Both GLM models had room for 8,000 tokens, or word pieces, per answer, and the note below the table says why.
Made-up DOIs from GLM-5.3 Prime and the models beside it. Three runs of the same 30 claims on October 23, 2026.
| Model | Role | Run 1One run puts all 30 claims to the model once. The prompt and the settings stay the same in each run. | Run 2 | Run 3 | Run spreadThe lowest run and the highest run. A wide spread tells you that one run alone can give a different answer. | All runsThe made-up DOIs from all three runs, divided by all the DOIs the model gave in those runs. That is about 135 DOIs for each model. | Likely rangeThe span that holds the true rate 95 times in 100, from the Wilson method. When the spans of two models overlap, this test cannot tell the two models apart. |
|---|---|---|---|---|---|---|---|
| GLM-5.3 Prime | subject | 37.8% | 26.7% | 24.4% | 24.4–37.8% | 29.6% | 22.6–37.8% |
| GLM-5.3 | predecessor | 27.9% | 27.9% | 41.5% | 27.9–41.5% | 31.5% | 24.2–40.0% |
| Grok 4.3 | control | 9.5% | 9.5% | 7.1% | 7.1–9.5% | 8.7% | 4.9–15.0% |
Each row shows one model. Run 1, Run 2 and Run 3 show the share of its DOIs that no registry knows. Run spread shows the lowest and the highest run. All runs adds the three runs together. Likely range shows where the true rate most likely sits, 95 times in 100.
How to read this
- Run
- One run puts all 30 claims to the model once. The prompt and the settings stay the same in each run.
- Run spread
- The lowest run and the highest run. A wide spread tells you that one run alone can give a different answer.
- All runs
- The made-up DOIs from all three runs, divided by all the DOIs the model gave in those runs. That is about 135 DOIs for each model.
- Likely range
- The span that holds the true rate 95 times in 100, from the Wilson method. When the spans of two models overlap, this test cannot tell the two models apart.
- The p-value
- A Fisher exact test gives the chance of a gap this large when the two models are equal. We treat a p-value above 0.05 as noise.
- Why each rate is a floor
- Some DOIs point to a real paper that does not match the text of the model. We do not count these DOIs as made up. So the true rate is equal to the rate here or higher.
The rate divides made-up DOIs by DOIs offered. A DOI counts as made up only when no registry has issued it. Pooled over three runs, GLM-5.3 Prime made up 40 of 135, with a 95 percent interval of 22.6 to 37.8 percent. GLM-5.3 made up 41 of 130, from 24.2 to 40.0 percent, and Grok 4.3 made up 11 of 126, from 4.9 to 15.0 percent. Fisher exact on the pair gives p = 0.79, and run by run it gives 0.37, 1.0 and 0.17. Another 67 DOIs from the pair led to a real paper whose title did not match what the model wrote. We graded each against the paper's registered first author and year, and 63 were miscitations: 24 from Prime and 39 from GLM-5.3. Three more DOIs from GLM-5.3 could not be looked up, so they sit in neither column, and its wrong count is out of 127. We did not read them by hand. Counting both kinds of error, Prime got 64 of 135 wrong, from 39.2 to 55.8 percent, and GLM-5.3 got 80 of 127, from 54.3 to 70.9 percent, p = 0.013. An answer that runs out of room comes back empty. In the first part of the test, at 3,000 tokens, GLM-5.3 ran out on 1 answer, so we ran the pair again at 8,000. GLM-5.3 still ran out once, on the claim about language models agreeing with users, in run 3. That answer came back empty and holds no DOI, so it does not enter the rate. Prime never ran out. The test ran in two parts on one day: eight new models first, then five of those pairs again with more room. The second part cost $3.88, and both parts together cost $11.65.
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.
Seven of its made-up IDs came back in every run
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. Neither GLM model offered a source for any of the fifteen in any run, and Prime gave sources for every real claim.
All 40 of Prime's made-up IDs came on real claims. Seven of them came back in all three runs, so asking it again will not catch them.
The 1994 Scandinavian Simvastatin Survival Study in The Lancet is real, and its ID is 10.1016/S0140-6736(94)90566-5. In all three runs, Prime dropped the S and gave 10.1016/0140-6736(94)90566-5, which points at nothing. GLM-5.3 made the same slip in run 3.
In all three runs, Prime cited Jolliffe's 2021 vitamin D trial as 10.1016/S2213-8587(21)00151-6. GLM-5.3 gave the same string in run 1 and 10.1016/S2213-8587(21)00251-6 in run 2. Neither ID exists.
In run 3, GLM-5.3 cited Doran's 2001 study of attention during sleep loss as 10.4436/0000-0000. That is an ID made of zeros, and no registry has issued it.
Three labs, one missing letter
The 4S statin trial has one ID, 10.1016/S0140-6736(94)90566-5, and GPT-6.1 Sol Pro gave it right in every run on 23 October. The same day, four models dropped the S and wrote the same dead string. They were GLM-5.3 Prime and GLM-5.3 from Z.ai, Muse Spark Open from Meta and GPT-6 Sol Pro from OpenAI. Models from three labs agreed on one slip. A registry lookup is the only check that settles whether an ID exists.
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.
GLM-5.3 Prime got about half its citations wrong, and seven of its fakes came back in every run. 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
GLM-5.3 Prime became available through the API on 23 September 2026. We measured it and put this page up on 23 October, 30 days later. That misses the 48-hour target we set ourselves by four weeks. We measured GLM-5.3 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. A model's rate can move between days, which is why we measure the old model again next to the new one. The combined gap clears the bar on that one day, and a second day for this pair is open work. Second, the two models are listed as the same model at different speeds, so a real gap between them has no known mechanism. Prime's true rate of made-up IDs could sit anywhere from about 23 to 38 percent, and GLM-5.3's from about 24 to 40. The pair also had more room per answer than the control: 8,000 tokens against 3,000 for Grok 4.3. Third, all the models ran at low reasoning effort with no web access. This test did not try a higher effort, web search or a sycophancy check, and we did not read the miscitations by hand. Our scorer checks DOIs against Crossref and DataCite. We then checked all 260 dead IDs from the second part of the 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 GLM-5.3 Prime hallucinate citations?
Yes, often. With no web access, it made up 40 of the 135 DOIs it offered across three runs on 23 October, or 29.6 percent. Counting real IDs on the wrong paper too, it got 47.4 percent of its citations wrong.
Is GLM-5.3 Prime better than GLM-5.3 at citations?
On made-up IDs, we found no difference: 29.6 percent against 31.5 (p = 0.79). On wrong citations overall, Prime was lower: 47.4 percent against 63.0 (p = 0.013). That gap comes from fewer miscitations, on one day, with no known cause.
Why would a faster version of the same model cite differently?
We do not know that it does. The two are listed as the same model. One day of runs showed fewer miscitations from Prime. A gap like that can come from chance, since we test many pairs, or from a change we cannot see. A second day would show which.
Can I trust GLM-5.3 Prime's citations without checking them?
No. About half its citations were wrong, and seven of its fakes came back in every run. Check each DOI against a registry before you cite it.
Why does another benchmark give GLM-5.3 Prime 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 part of the test behind this page cost $3.88, and the whole test cost $11.65. Ask and we will send the run files.
Is this the GLM-5.3 Prime 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.