# o1-preview ![Organization](https://img.shields.io/badge/Org-OpenAI-74aa9c) ![Context](https://img.shields.io/badge/Context-128K%20tokens-1f6feb) ![License](https://img.shields.io/badge/License-Proprietary-d29922) ![Released](https://img.shields.io/badge/Released-Sep%202024-555) ![Reasoning](https://img.shields.io/badge/Type-Reasoning-8250df) > **Quick answer:** o1-preview is OpenAI's September 2024 first public reasoning model, scoring 85.5% on MATH, 90.8% on MMLU, and 42.4% on SimpleQA. It introduced chain-of-thought reasoning to the OpenAI API — a preview release ahead of the full o1 model. It has been superseded by o1, o3, and o4-mini. ## At a Glance **Where o1-preview leads** - First public o-series reasoning model — historically significant - 90.8% MMLU — strong general knowledge (higher than o3-mini's 86.9%) - 42.4% SimpleQA — better factual accuracy than later o-series models - 85.5% MATH — strong mathematical reasoning for a 2024 model - Good factual recall relative to other reasoning models **Where it lags** - Preview model — limited features vs full o1 - Superseded by o1, o3-mini, o3, and o4-mini in all dimensions - October 2023 knowledge cutoff — over a year behind current models - No vision support - 128K context (vs 200K for o3-mini and o4-mini) - Pricing no longer publicly listed **Best for:** Historical comparisons; legacy integrations; understanding o-series model progression. ## What o1-preview Is o1-preview (released September 12, 2024) was OpenAI's first public deployment of the o-series reasoning paradigm — making chain-of-thought reasoning available through the API before the full o1 model launched in December 2024. The model introduced a fundamentally different approach to generation: instead of producing output directly, o-series models "think" through problems using internal reasoning steps, then produce a final answer. This yielded dramatically improved performance on mathematics (85.5% MATH), science, and coding compared to GPT-4o. o1-preview was explicitly a preview — limited context (128K vs 200K for o1), no image support, and fewer system prompt capabilities than the production o1. It was retired as the o-series portfolio expanded. For historical context: o1-preview demonstrated that reasoning models would become the standard for complex tasks, setting the stage for the entire o-series roadmap. ## Specifications | Field | Value | |---|---| | Organization | OpenAI | | Context window | 128,000 tokens | | License | Proprietary (API only) | | Release date | September 12, 2024 | | Knowledge cutoff | October 2023 | | Modality | Text only | ## Pricing o1-preview pricing is no longer publicly listed on the OpenAI pricing page — this model has been deprecated in favor of o1, o3-mini, and o4-mini. ## Public Benchmark Scores | Benchmark | Score | Source | Date | |---|---|---|---| | [MATH](/benchmarks/math) | 85.5% | Benchgen evaluation | 2025-07 | | [MMLU](/benchmarks/mmlu) | 90.8% | Benchgen evaluation | 2025-07 | | [SimpleQA](/benchmarks/simpleqa) | 42.4% | Benchgen evaluation | 2025-07 | ## o1-preview vs Alternatives | Model | MATH | MMLU | SimpleQA | Context | |---|---|---|---|---| | **o1-preview** | 85.5% | 90.8% | 42.4% | 128K | | [o3-mini](/models/openai/o3-mini) | 97.9% | 86.9% | 15.0% | 200K | | [o1](/models/openai/o1) | — | — | — | 200K | | [o4-mini](/models/openai/o4-mini) | — | — | — | 200K | o1-preview vs o3-mini: o3-mini is much stronger on MATH (97.9% vs 85.5%), but o1-preview has higher SimpleQA (42.4% vs 15.0%). o1-preview is a legacy model; use o3-mini or o4-mini for new work. ## Related Models & Benchmarks - Compare: [o3-mini](/models/openai/o3-mini) · [o4-mini](/models/openai/o4-mini) · [o1](/models/openai/o1) - Benchmarks: [MATH](/benchmarks/math) · [MMLU](/benchmarks/mmlu) · [SimpleQA](/benchmarks/simpleqa) ## Frequently Asked Questions

What is o1-preview?

o1-preview is OpenAI's September 2024 first public reasoning model, scoring 85.5% MATH, 90.8% MMLU, and 42.4% SimpleQA. It was a preview release ahead of the full o1 model and has been superseded.

Is o1-preview still available?

o1-preview has been deprecated and is no longer listed on the OpenAI pricing page. Migrate to o1, o3-mini, or o4-mini.

What is the difference between o1-preview and o1?

o1-preview was a limited preview with 128K context and restricted features. The full o1 (December 2024) added 200K context, vision support, and more complete system prompt capabilities.

--- *Specs from OpenAI's o1-preview launch (September 2024) and Benchgen evaluations. Last updated 2026-07-24.*

o1-preview — Evaluations

Evaluations