Atlas

Models

← All models

gpt-oss-120b

OpenAI · 2025-08-05

gpt-oss-120b is the larger of OpenAI's two Apache 2.0-licensed gpt-oss reasoning models, which are the company's first open-weight language models since GPT-2. It is a text-only mixture-of-experts model with approximately 116.8B total parameters and 5.1B active parameters per token, configurable reasoning effort, and training for tool use.

Benchmark scores

BenchmarkScore
AA-Briefcase Elo7.0
AA-LCR50.7
AA-Omniscience Index-50.0
Aider Polyglot41.8
AIME 2024 (avg@8)93.1
AIME 2024-202592.6
AIME 202566.7
AIME 2025 (avg@8)92.1
ALE-Bench1119
AlgoTune1.4
Apex (MathArena)1.0
Apex Shortlist45.2
APEX-Agents4.7
APEX-Agents-AA3.1
Arena Text — Business, Management & Financial Ops (No Style Control)1358
Arena Text — Chinese (No Style Control)1377
Arena Text — Creative Writing (No Style Control)1273
Arena Text — English (No Style Control)1374
Arena Text — Entertainment, Sports & Media (No Style Control)1293
Arena Text — Exclude Ties (No Style Control)1331
Arena Text — Exclude Ties (Style Controlled)1314
Arena Text — Expert (No Style Control)1357
Arena Text — French (No Style Control)1370
Arena Text — French (Style Controlled)1353
Arena Text — German (No Style Control)1352
Arena Text — Hard Prompts (No Style Control)1364
Arena Text — Hard Prompts English (No Style Control)1368
Arena Text — Instruction Following (No Style Control)1319
Arena Text — Japanese (No Style Control)1324
Arena Text — Korean (No Style Control)1280
Loading Atlas data…