Name: GPT-3.5 Turbo 0125
Author: OpenAI

GPT-3.5 Turbo 0125

OpenAI · 2024-02-01

GPT-3.5 Turbo 0125 is the GPT-3.5 Turbo chat-model snapshot that OpenAI released on February 1, 2024. OpenAI described it as improving accuracy when responding in requested formats and fixing a text-encoding bug affecting non-English function calls. OpenAI has not publicly disclosed its parameter count.

Benchmark scores

Benchmark	Score
Arena Text — Business, Management & Financial Ops (No Style Control)	1076
Arena Text — Chinese (No Style Control)	1074
Arena Text — Creative Writing (No Style Control)	1093
Arena Text — English (No Style Control)	1139
Arena Text — Entertainment, Sports & Media (No Style Control)	1089
Arena Text — Exclude Ties (No Style Control)	968
Arena Text — Exclude Ties (Style Controlled)	1111
Arena Text — Expert (No Style Control)	1065
Arena Text — French (No Style Control)	1116
Arena Text — French (Style Controlled)	1230
Arena Text — German (No Style Control)	1090
Arena Text — Hard Prompts (No Style Control)	1108
Arena Text — Hard Prompts English (No Style Control)	1124
Arena Text — Instruction Following (No Style Control)	1119
Arena Text — Japanese (No Style Control)	1044
Arena Text — Korean (No Style Control)	1019
Arena Text — Legal & Government (No Style Control)	1118
Arena Text — Life, Physical & Social Science (No Style Control)	1087
Arena Text — Longer Query (No Style Control)	1121
Arena Text — Medicine & Healthcare (No Style Control)	1073
Arena Text — Multi-turn (No Style Control)	1117
Arena Text — Non-English (No Style Control)	1107
Arena Text — Non-English (Style Controlled)	1207
Arena Text — Occupational Mathematical (No Style Control)	1127
Arena Text — Polish (No Style Control)	1120
Arena Text — Polish (Style Controlled)	1226
Arena Text — Russian (No Style Control)	1122
Arena Text — Software & IT Services (No Style Control)	1116
Arena Text — Spanish (No Style Control)	1120
Arena Text — Writing, Literature & Language (No Style Control)	1139