Atlas

Models

← All models

O1

OpenAI · 2024-12-05

OpenAI o1 is the production release of OpenAI's first o-series reasoning model, made generally available on December 5, 2024 after the earlier September o1-preview. It adds multimodal image input, a 200,000-token context window, function calling, developer messages, Structured Outputs, and a reasoning_effort parameter in the API, while using fewer reasoning tokens than o1-preview for comparable requests.

Benchmark scores

BenchmarkScore
AA-LCR59.3
AA-Omniscience Index-10.6
Agentic Tool Use (Chat)60.4
Agentic Tool Use (Enterprise)70.1
AidanBench6057
Aider Polyglot61.7
AIME 2024 (avg@8)73.3
AIME 2024-202571.5
AIME 202581.7
AIME 2025 (avg@8)69.6
APEX-Agents1.1
Arabic1120
ARC-AGI-130.7
Artificial Analysis Coding Index39.7
Artificial Analysis Intelligence Index23.4
Artificial Analysis Omniscience Accuracy34.7
Artificial Analysis Omniscience Hallucination Rate69.3
BioTIER Refusal64.6
BrowseComp9.9
BuseyBench SVG4.1
BuzzBench59.3
CadEval56.0
CadQueryEval44.0
CadQueryEval — Bounding Box48.0
CadQueryEval — Chamfer Distance48.0
CadQueryEval — Code Execution52.0
CadQueryEval — Component Count48.0
CadQueryEval — Hausdorff 95p48.0
CadQueryEval — STL Creation52.0
CadQueryEval — Volume44.0
Loading Atlas data…