Atlas

Models

← All models

OLMo 2 32B Instruct

Allen Institute for AI (Ai2) · 2025-03-13 · 32.2B parameters

OLMo 2 32B Instruct is Ai2's fully open post-trained variant of OLMo 2 32B, produced through supervised fine-tuning, DPO, and final reinforcement learning with verifiable rewards using a Tülu 3.1-style recipe. Ai2 released its weights, training data and code, logs, and intermediate RLVR checkpoints.

Benchmark scores

BenchmarkScore
Arena Text — Business, Management & Financial Ops (No Style Control)1181
Arena Text — Chinese (No Style Control)1191
Arena Text — Creative Writing (No Style Control)1199
Arena Text — English (No Style Control)1255
Arena Text — Entertainment, Sports & Media (No Style Control)1225
Arena Text — Exclude Ties (No Style Control)1107
Arena Text — Exclude Ties (Style Controlled)1154
Arena Text — Hard Prompts (No Style Control)1208
Arena Text — Hard Prompts English (No Style Control)1244
Arena Text — Instruction Following (No Style Control)1187
Arena Text — Legal & Government (No Style Control)1228
Arena Text — Life, Physical & Social Science (No Style Control)1240
Arena Text — Longer Query (No Style Control)1194
Arena Text — Medicine & Healthcare (No Style Control)1181
Arena Text — Multi-turn (No Style Control)1221
Arena Text — Non-English (No Style Control)1161
Arena Text — Non-English (Style Controlled)1196
Arena Text — Occupational Mathematical (No Style Control)1195
Arena Text — Russian (No Style Control)1186
Arena Text — Software & IT Services (No Style Control)1215
Arena Text — Writing, Literature & Language (No Style Control)1204
LMArena Text Arena Chinese Elo1240
LMArena Text Arena Coding Elo1280
LMArena Text Arena Creative Writing Elo1224
LMArena Text Arena Elo1251
LMArena Text Arena English Elo1287
LMArena Text Arena Hard Prompts Elo1261
LMArena Text Arena Hard Prompts English Elo1296
LMArena Text Arena Industry: Writing, Literature & Language Elo1233
LMArena Text Arena Instruction Following Elo1229
Loading Atlas data…