Closed SourceOpenAIReleased 2024-09

o1

OpenAI's reasoning model that thinks before it answers - best for hard science and math

Context window

200K

Input / 1M tokens

$15.00

Output / 1M tokens

$60.00

Provider

OpenAI

o1 uses chain-of-thought reasoning internally before producing a response, spending extra compute on thinking rather than output tokens. This approach makes it dramatically better than GPT-4o on hard math, science, and complex coding problems. The trade-off is higher latency and cost - o1 is best used for problems where quality matters more than speed.

Strengths

  • Best-in-class math and physics
  • Strong competitive coding (Codeforces, HumanEval)
  • Scientific reasoning (GPQA top performer)
  • Multi-step logic and planning
  • 200K context for long technical documents

Best for developers who...

Math and science problemsCompetitive programmingComplex multi-step reasoningResearch assistance

Benchmarks

BenchmarkScoreNotes
GPQA Diamond78.3%Expert-level science questions
HumanEval92.4%Near-perfect coding
SWE-bench Verified48.9%Strong software engineering

Source: OpenAI o1 technical overview

Compare o1 with

All model comparisons →

Learn the concepts