List your company

Machine Translation on FLORES-200 (devtest)

5.67Delta CHRF++SFT 3ep
-0.12281.38112.8854.3889May 15, 2026

Evaluation results

MethodLinks
SFT 3epModel Parameters=600M, Language=zho_Hant, Language Group=Isol.2026.055.67——————0.10.138——
GRPOModel Parameters=600M, Language=zho_Hant, Language Group=Isol.2026.055.07——————0.0920.115——
SFT 1epModel Parameters=600M, Language=zho_Hant, Language Group=Isol.2026.053.74——————0.0690.102——
GRPOModel Parameters=600M, Language=bod, Language Group=Aggl.2026.053.01——————0.0040.012——
GRPOModel Parameters=600M, Language=eus, Language Group=Aggl.2026.052.93——————0.0120.006——
SFT 3epModel Parameters=600M, Language=zho_Hans, Language Group=Isol.2026.052.88——————0.0390.071——
SFT 3epModel Parameters=600M, Language=bod, Language Group=Aggl.2026.052.85——————0.0680.032——
SFT 3epModel Parameters=600M, Language=swh, Language Group=Aggl.2026.052.7——————0.0080.011——
SFT 3epModel Parameters=600M, Language=eus, Language Group=Aggl.2026.052.48——————0.0110.005——
SFT 1epModel Parameters=600M, Language=zho_Hans, Language Group=Isol.2026.052.29——————0.030.062——
GRPOModel Parameters=600M, Language=zho_Hans, Language Group=Isol.2026.052.28——————0.0340.04——
SFT 3epModel Parameters=600M, Language=ben, Language Group=Fus.2026.052.15——————0.0070.007——
GRPOModel Parameters=600M, Language=ben, Language Group=Fus.2026.051.99——————0.0040.006——
GRPOModel Parameters=600M, Language=swh, Language Group=Aggl.2026.051.88——————0.0040.008——
SFT 3epModel Parameters=600M, Language=tur, Language Group=Aggl.2026.051.8——————0.0050.005——
SFT 3epModel Parameters=600M, Language=yor, Language Group=Isol.2026.051.56——————0.0220.001——
SFT 3epModel Parameters=600M, Language=jpn, Language Group=Aggl.2026.051.46——————0.0060.007——
SFT 1epModel Parameters=600M, Language=bod, Language Group=Aggl.2026.051.43——————0.0350.017——
SFT 1epModel Parameters=600M, Language=yor, Language Group=Isol.2026.051.28——————0.0110——
SFT 3epModel Parameters=600M, Language=arb, Language Group=Root-pat.2026.051.27——————0.0060.005——
SFT 1epModel Parameters=600M, Language=swh, Language Group=Aggl.2026.051.18——————0.0030.005——
SFT 1epModel Parameters=600M, Language=eus, Language Group=Aggl.2026.051.1——————0.0050.003——
SFT 3epModel Parameters=600M, Language=ces, Language Group=Fus.2026.051.08——————0.0050.003——
SFT 1epModel Parameters=600M, Language=ben, Language Group=Fus.2026.051.07——————0.0030.003——
GRPOModel Parameters=600M, Language=yor, Language Group=Isol.2026.050.97——————0.0040.007——
SFT 1epModel Parameters=600M, Language=jpn, Language Group=Aggl.2026.050.96——————0.0120.009——
GRPOModel Parameters=600M, Language=tur, Language Group=Aggl.2026.050.95——————0.0010.002——
GRPOModel Parameters=600M, Language=jpn, Language Group=Aggl.2026.050.86——————0.0180.011——
SFT 1epModel Parameters=600M, Language=tur, Language Group=Aggl.2026.050.68——————0.0020.002——
SFT 3epModel Parameters=600M, Language=pol, Language Group=Fus.2026.050.68——————0.0030.001——
GRPOModel Parameters=600M, Language=ces, Language Group=Fus.2026.050.57——————0.0010.002——
GRPOModel Parameters=600M, Language=bel, Language Group=Fus.2026.050.55——————0.0080——
GRPOModel Parameters=600M, Language=pol, Language Group=Fus.2026.050.55——————00.001——
SFT 3epModel Parameters=600M, Language=bel, Language Group=Fus.2026.050.43——————0.0070——
SFT 1epModel Parameters=600M, Language=arb, Language Group=Root-pat.2026.050.41——————0.0040.002——
SFT 1epModel Parameters=600M, Language=pol, Language Group=Fus.2026.050.27——————0.0020.001——
SFT 1epModel Parameters=600M, Language=ces, Language Group=Fus.2026.050.17——————0.0020.001——
GRPOModel Parameters=600M, Language=arb, Language Group=Root-pat.2026.050.1——————00.001——
SFT 1epModel Parameters=600M, Language=bel, Language Group=Fus.2026.050.1——————0.0030——
BaseModel Parameters=600M, Language=eus, Language Group=Aggl.2026.05————52.69——————
BaseModel Parameters=600M, Language=swh, Language Group=Aggl.2026.05————60.33——————
BaseModel Parameters=600M, Language=bod, Language Group=Aggl.2026.05————23.37——————
BaseModel Parameters=600M, Language=tur, Language Group=Aggl.2026.05————56.49——————
BaseModel Parameters=600M, Language=jpn, Language Group=Aggl.2026.05————29.31——————
BaseModel Parameters=600M, Language=zho_Hant, Language Group=Isol.2026.05————16.6——————
BaseModel Parameters=600M, Language=zho_Hans, Language Group=Isol.2026.05————24.59——————
BaseModel Parameters=600M, Language=yor, Language Group=Isol.2026.05————24.28——————
BaseModel Parameters=600M, Language=arb, Language Group=Root-pat.2026.05————54.72——————
BaseModel Parameters=600M, Language=bel, Language Group=Fus.2026.05————41.75——————
BaseModel Parameters=600M, Language=ben, Language Group=Fus.2026.05————49.84——————
BaseModel Parameters=600M, Language=ces, Language Group=Fus.2026.05————55.13——————
BaseModel Parameters=600M, Language=pol, Language Group=Fus.2026.05————48.31——————
Final MarianMT Model2026.02—29.04——58.14—0.6314————
GEMMA-3-12B-PTshots=5, averaging=across all studied translation directions2026.02—37.514.063.6355.370.84—————
GEMMA-3-1B-PTshots=5, averaging=across all studied translation directions2026.02—24.557.296.5944.120.71—————
GEMMA-3-270Mshots=5, averaging=across all studied translation directions2026.02—12.1513.5512.7930.660.39—————
GEMMA-3-4B-PTshots=5, averaging=across all studied translation directions2026.02—33.154.794.2551.840.82—————
GPT-OSS-20Bk=–, Expert drop=0%, Params=20.9B, Recovery fine-tuning=None, Calibration direction=N/A2026.05—————0.942———0.9260.958
LLAMA-3.2-1Bshots=5, averaging=across all studied translation directions2026.02—19.349.278.4838.580.59—————
LLAMA-3.2-3Bshots=5, averaging=across all studied translation directions2026.02—26.746.555.7846.020.73—————
Multiling. En→Xk=16, Expert drop=50.00%, Params=11.3B, Recovery fine-tuning=no retrain, Calibration direction=English→X2026.05—————0.93———0.9090.951
Multiling. En→Xk=20, Expert drop=62.50%, Params=9.0B, Recovery fine-tuning=no retrain, Calibration direction=English→X2026.05—————0.896———0.8630.929
Multiling. En→Xk=22, Expert drop=68.75%, Params=7.8B, Recovery fine-tuning=no retrain, Calibration direction=English→X2026.05—————0.859———0.8190.899
Multiling. En→Xk=22, Expert drop=68.75%, Params=7.8B, Recovery fine-tuning=10k distil., Calibration direction=English→X2026.05—————0.904———0.8740.934
Multiling. En→Xk=24, Expert drop=75.00%, Params=6.6B, Recovery fine-tuning=10k distil., Calibration direction=English→X2026.05—————0.902———0.8690.936
Multiling. En→Xk=26, Expert drop=81.25%, Params=5.4B, Recovery fine-tuning=10k distil., Calibration direction=English→X2026.05—————0.89———0.8470.932
Multiling. En→Xk=28, Expert drop=87.50%, Params=4.2B, Recovery fine-tuning=10k distil., Calibration direction=English→X2026.05—————0.871———0.8120.93
NLLB-200k=–, Expert drop=–, Params=3.3B, Recovery fine-tuning=None, Calibration direction=N/A2026.05—————0.918———0.8820.954
Previous Best2026.02—20.76—————————
QWEN3-0.6B-BASEshots=5, averaging=across all studied translation directions2026.02—17.4710.839.8736.360.55—————
QWEN3-1.7B-BASEshots=5, averaging=across all studied translation directions2026.02—23.648.157.3442.410.67—————
QWEN3-4B-BASEshots=5, averaging=across all studied translation directions2026.02—28.166.715.9846.740.73—————
TranslateGemma-4Bk=–, Expert drop=–, Params=4.3B, Recovery fine-tuning=None, Calibration direction=N/A2026.05—————0.945———0.9330.958