Benchmark
Benchmark results
Document scores
Average error = how many points the detector's AI percentage was off from Turnitin's, averaged over every document. Lower is better.
| Detector | Docs scored | Avg. error | Median error | False flags* | Missed AI* |
|---|---|---|---|---|---|
| academi.cx | 429 | 1.9 pts | 0 pts | 0% | 0% |
| Copyleaks | 129 | 8.2 pts | 0 pts | 8.8% | 7.7% |
| Winston AI | 429 | 9.4 pts | 1 pts | 14.5% | 7.1% |
| Originality.ai | 429 | 12.9 pts | 5 pts | 24.1% | 7.1% |
| Pangram | 429 | 13.1 pts | 0 pts | 11.4% | 6.1% |
| GPTZero | 429 | 20.5 pts | 3.77 pts | 13.2% | 33.3% |
| ZeroGPT | 429 | 22.2 pts | 14.4 pts | 20.5% | 66.7% |
*False flags: human-written documents scored above 20%. Missed AI: fully AI-written documents scored below 80%. Copyleaks was only tested on 129 documents due to limits.
By document type
Average error against Turnitin's score, categorized by Turnitin's returned percentage on the document. Lower is better.
| Detector | Scored 0% AI on Turnitin | Scored 20–99% AI on Turnitin | Scored 100% AI on Turnitin |
|---|---|---|---|
| academi.cx | 0.2 pts (220 docs) | 4 pts (110 docs) | 3.5 pts (99 docs) |
| Copyleaks | 5.7 pts (80 docs) | 17.6 pts (23 docs) | 7.7 pts (26 docs) |
| Winston AI | 9.2 pts (220 docs) | 14.7 pts (110 docs) | 4 pts (99 docs) |
| Originality.ai | 14.5 pts (220 docs) | 17.1 pts (110 docs) | 4.8 pts (99 docs) |
| Pangram | 7.6 pts (220 docs) | 32 pts (110 docs) | 4.2 pts (99 docs) |
| GPTZero | 9.3 pts (220 docs) | 36.3 pts (110 docs) | 27.8 pts (99 docs) |
| ZeroGPT | 11.9 pts (220 docs) | 27.1 pts (110 docs) | 39.9 pts (99 docs) |
Detector scores vs. Turnitin
Each dot is one of the 429 documents in the benchmark. Tighter plotting against the dashed line is better.
Turnitin's score runs across, the detector's score runs up. Click a plot to enlarge it.
Sentence accuracy
A sentence counts as correct when the detector and Turnitin agree on AI vs. human. Pangram has been excluded as they provide no sentence-specific data, only extremely large windows, which makes fair comparison impossible.
| Detector | Sentences scored | Classified correctly | Accuracy |
|---|---|---|---|
| academi.cx | 57,478 | 55,547 | 96.6% |
| Copyleaks | 13,467 | 11,469 | 85.2% |
| Winston AI | 56,559 | 43,319 | 76.6% |
| Originality.ai | 57,478 | 43,561 | 75.8% |
| GPTZero | 49,079 | 35,668 | 72.7% |
| ZeroGPT | 54,619 | 39,154 | 71.7% |
Raw results
| Doc | Words | Turnitin | academi.cx | Copyleaks | Winston AI | Originality.ai | Pangram | GPTZero | ZeroGPT |
|---|---|---|---|---|---|---|---|---|---|
| 0%_1 | 1,355 | 0% | 0% (0) | — | 0% (0) | 0.5% (0.5) | 0% (0) | 0.22% (0.2) | 6% (6) |
| 0%_2 | 2,027 | 0% | 0% (0) | — | 0% (0) | 1.21% (1.2) | 0% (0) | 6.81% (6.8) | 3.3% (3.3) |
| 0%_3 | 1,065 | 0% | 0% (0) | 0% (0) | 0.17% (0.2) | 2.25% (2.3) | 0% (0) | 0.27% (0.3) | 4% (4) |
| 0%_4 | 2,755 | 0% | 0% (0) | 0% (0) | 3.41% (3.4) | 1.93% (1.9) | 0% (0) | 0% (0) | 9.1% (9.1) |
| 0%_5 | 1,773 | 0% | 0% (0) | 0% (0) | 0.06% (0.1) | 0.79% (0.8) | 0% (0) | 2.13% (2.1) | 8.6% (8.6) |
| 0%_6 | 504 | 0% | 0% (0) | 0% (0) | 0.38% (0.4) | 0.29% (0.3) | 0.05% (0.1) | 0.04% (0) | 3% (3) |
| 0%_7 | 1,177 | 0% | 0% (0) | 0% (0) | 3.88% (3.9) | 0.39% (0.4) | 0% (0) | 0.01% (0) | 3.7% (3.7) |
| 0%_8 | 2,320 | 0% | 0% (0) | 0% (0) | 3.41% (3.4) | 6.27% (6.3) | 0% (0) | 0% (0) | 6.7% (6.7) |
| 0%_9 | 1,006 | 0% | 0% (0) | 0% (0) | 0% (0) | 2.35% (2.4) | 0% (0) | 4.23% (4.2) | 11.9% (11.9) |
| 0%_10 | 596 | 0% | 0% (0) | 0% (0) | 0.01% (0) | 0.29% (0.3) | 0% (0) | 0% (0) | 20.4% (20.4) |
| 0%_11 | 2,138 | 0% | 0% (0) | 20.91% (20.9) | 0.17% (0.2) | 1.51% (1.5) | 0.02% (0) | 5.63% (5.6) | 0% (0) |
| 0%_12 | 5,685 | 0% | 0% (0) | 5.22% (5.2) | 1.5% (1.5) | 2.39% (2.4) | 0% (0) | 12.77% (12.8) | 5.4% (5.4) |
| 0%_13 | 966 | 0% | 0% (0) | 0% (0) | 6.37% (6.4) | 24.16% (24.2) | 0.01% (0) | 0.01% (0) | 15.8% (15.8) |
| 0%_14 | 640 | 0% | 0% (0) | 0% (0) | 0% (0) | 0.89% (0.9) | 0% (0) | 0.21% (0.2) | 0% (0) |
| 0%_15 | 3,533 | 0% | 0% (0) | 6.17% (6.2) | 9.32% (9.3) | 1.78% (1.8) | 0% (0) | 12% (12) | 3.3% (3.3) |
| 0%_16 | 2,019 | 0% | 2% (2) | 0% (0) | 22.69% (22.7) | 2.93% (2.9) | 0% (0) | 0.47% (0.5) | 14.4% (14.4) |
| 0%_17 | 687 | 0% | 0% (0) | 0% (0) | 0.15% (0.2) | 45% (45) | 57.76% (57.8) | 0.01% (0) | 0% (0) |
| 0%_18 | 20,680 | 0% | 0% (0) | 0% (0) | 0.01% (0) | 1.35% (1.4) | 0% (0) | 0% (0) | 7.1% (7.1) |
| 0%_19 | 834 | 0% | 0% (0) | 0% (0) | 0.12% (0.1) | 0.71% (0.7) | 0% (0) | 0.3% (0.3) | 8.6% (8.6) |
| 0%_20 | 883 | 0% | 0% (0) | 0% (0) | 0% (0) | 0.25% (0.3) | 0% (0) | 0% (0) | 0% (0) |
| 0%_21 | 712 | 0% | 0% (0) | 0% (0) | 0% (0) | 0.6% (0.6) | 0% (0) | 0% (0) | 11.2% (11.2) |
| 0%_22 | 2,537 | 0% | 0% (0) | 0% (0) | 0.65% (0.7) | 39.21% (39.2) | 0% (0) | 0% (0) | 6% (6) |
| 0%_23 | 548 | 0% | 0% (0) | 0% (0) | 0% (0) | 22.55% (22.6) | 2.09% (2.1) | 22.31% (22.3) | 0% (0) |
| 0%_24 | 5,746 | 0% | 0% (0) | 0% (0) | 0.14% (0.1) | 44.87% (44.9) | 0% (0) | 0.07% (0.1) | 22.1% (22.1) |
| 0%_25 | 1,063 | 0% | 0% (0) | 100% (100) | 26.66% (26.7) | 99.89% (99.9) | 0% (0) | 0.02% (0) | 53.4% (53.4) |
(n) = points off Turnitin's score · — = not scored