bigscience·released 2022-10-081 source
bloomz-560m benchmark scores: 53 benchmarks tracked, leading 51. 0% of its scores are independently reproduced — the rest are self-reported or unverified.
Leads on 51 of 51 benchmarks·best result 67.15 on SuperGLUE (rte) (unverified)·0 of 53 independently reproduced
Leads on: ANLI (r1), ANLI (r2), ANLI (r3), HumanEval, StoryCloze (2016), SuperGLUE (cb), SuperGLUE (copa), SuperGLUE (rte), Winogrande XL (xl), XCOPA (et), XCOPA (ht), XCOPA (id), XCOPA (it), XCOPA (qu), XCOPA (sw), XCOPA (ta), XCOPA (th), XCOPA (tr), XCOPA (vi), XCOPA (zh), XNLI (ar), XNLI (bg), XNLI (de), XNLI (el), XNLI (en), XNLI (es), XNLI (fr), XNLI (hi), XNLI (ru), XNLI (sw), XNLI (th), XNLI (tr), XNLI (ur), XNLI (vi), XNLI (zh), XStoryCloze (ar), XStoryCloze (es), XStoryCloze (eu), XStoryCloze (hi), XStoryCloze (id), XStoryCloze (my), XStoryCloze (ru), XStoryCloze (sw), XStoryCloze (te), XStoryCloze (zh), XWinograd (en), XWinograd (fr), XWinograd (jp), XWinograd (pt), XWinograd (ru), XWinograd (zh)
| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| SuperGLUE (rte) | 67.15Accuracy | unverifiedT2 | — |
| XCOPA (tr) | 61Accuracy | unverifiedT2 | — |
| XCOPA (vi) | 61Accuracy | unverifiedT2 | — |
| XCOPA (zh) | 61Accuracy | unverifiedT2 | — |
| StoryCloze (2016) | 60.29Accuracy | unverifiedT2 | — |
| XCOPA (ta) | 58Accuracy | unverifiedT2 | — |
| XCOPA (th) | 58Accuracy | unverifiedT2 | — |
| XStoryCloze (id) | 57.84Accuracy | unverifiedT2 | — |
| XCOPA (id) | 57Accuracy | unverifiedT2 | — |
| XStoryCloze (zh) | 56.52Accuracy | unverifiedT2 | — |
| XStoryCloze (es) | 56.45Accuracy | unverifiedT2 | — |
| XCOPA (sw) | 56Accuracy | unverifiedT2 | — |
| XStoryCloze (hi) | 55.79Accuracy | unverifiedT2 | — |
| XCOPA (qu) | 55Accuracy | unverifiedT2 | — |
| XStoryCloze (te) | 54.86Accuracy | unverifiedT2 | — |
| XWinograd (zh) | 54.76Accuracy | unverifiedT2 | — |
| XStoryCloze (ar) | 54.4Accuracy | unverifiedT2 | — |
| XWinograd (pt) | 53.99Accuracy | unverifiedT2 | — |
| XWinograd (ru) | 53.97Accuracy | unverifiedT2 | — |
| SuperGLUE (cb) | 53.57Accuracy | unverifiedT2 | — |
| XStoryCloze (ru) | 53.14Accuracy | unverifiedT2 | — |
| XCOPA (et) | 53Accuracy | unverifiedT2 | — |
| Winogrande XL (xl) | 52.41Accuracy | unverifiedT2 | — |
| XWinograd (jp) | 52.03Accuracy | unverifiedT2 | — |
| SuperGLUE (copa) | 52Accuracy | unverifiedT2 | — |
| XCOPA (it) | 52Accuracy | unverifiedT2 | — |
| XWinograd (fr) | 51.81Accuracy | unverifiedT2 | — |
| XStoryCloze (sw) | 51.36Accuracy | unverifiedT2 | — |
| XWinograd (en) | 51.01Accuracy | unverifiedT2 | — |
| XStoryCloze (eu) | 50.56Accuracy | unverifiedT2 | — |
| XCOPA (ht) | 49Accuracy | unverifiedT2 | — |
| XStoryCloze (my) | 47.05Accuracy | unverifiedT2 | — |
| XNLI (en) | 46.43Accuracy | unverifiedT2 | — |
| XNLI (fr) | 45.54Accuracy | unverifiedT2 | — |
| XNLI (es) | 44.98Accuracy | unverifiedT2 | — |
| XNLI (vi) | 44.74Accuracy | unverifiedT2 | — |
| XNLI (zh) | 44.66Accuracy | unverifiedT2 | — |
| XNLI (ar) | 44.46Accuracy | unverifiedT2 | — |
| XNLI (hi) | 41.81Accuracy | unverifiedT2 | — |
| XNLI (el) | 40.96Accuracy | unverifiedT2 | — |
| XNLI (bg) | 39.76Accuracy | unverifiedT2 | — |
| XNLI (ru) | 39.64Accuracy | unverifiedT2 | — |
| XNLI (de) | 39.36Accuracy | unverifiedT2 | — |
| XNLI (ur) | 38.96Accuracy | unverifiedT2 | — |
| XNLI (sw) | 38.35Accuracy | unverifiedT2 | — |
| XNLI (tr) | 37.31Accuracy | unverifiedT2 | — |
| XNLI (th) | 35.5Accuracy | unverifiedT2 | — |
| ANLI (r3) | 33.5Accuracy | unverifiedT2 | — |
| ANLI (r1) | 33.4Accuracy | unverifiedT2 | — |
| ANLI (r2) | 33.4Accuracy | unverifiedT2 | — |
| HumanEval | 9Pass@100 | unverifiedT2 | — |
| HumanEval | 4.11Pass@10 | unverifiedT2 | — |
| HumanEval | 2.18Pass@1 | unverifiedT2 | — |
Claims drawn from cited facts, not live model generation.
Developed by BigScience, this model is a compact 0.56-billion-parameter system released in 2022. Its compact scale favors efficiency and lower cost rather than the greater headroom and expense of larger models.
4 cited facts
This model is scored on 51 tracked benchmarks, and it currently holds the top score on all 51 of them. Its record is unverified: no outside evaluation has either confirmed or disputed the reported results.
3 cited facts
On average it sits 0.0 points off the best score across its benchmarks (a gap under a disclosed harness, not a capability claim), which falls in the couple-of-points-or-less band and reads as at the frontier. It holds the top score on 51 benchmarks, which is genuine category leadership rather than a near-miss.
2 cited facts
Benchmarks with more than one recorded measurement for this model — every one shown, not just the headline number.
2 facts cross-checked across data sources: 2 single-source.
Source facts, citations, and refresh stamp for this record.
Sources: huggingface_models
bloomz-560m is an AI model developed by bigscience, released 2022-10-08. Its benchmark record below tags every score as independently reproduced, self-reported, or unverified.
bloomz-560m has recorded scores on 51 benchmarks, each shown with its evidence status.
bloomz-560m has recorded scores on 53 benchmarks — Winogrande XL (xl), XWinograd (en), XWinograd (fr), XWinograd (jp), XWinograd (pt), XWinograd (ru), and 47 more. The full table above shows each score with its evidence status.
0 of 53 recorded scores (0%) are independently reproduced rather than self-reported by the lab.
bloomz-560m has 51 tracked claims: 51 unverified.