# o1 ![Organization](https://img.shields.io/badge/Org-OpenAI-412991) ![Context](https://img.shields.io/badge/Context-200K%20tokens-1f6feb) ![Pricing](https://img.shields.io/badge/Price-%2415%20%2F%20%2460%20per%201M-2da44e) ![License](https://img.shields.io/badge/License-Proprietary-d29922) ![Modality](https://img.shields.io/badge/Modality-Text%20%2B%20Vision-8957e5) ![Released](https://img.shields.io/badge/Released-Dec%202024-555) > **Quick answer:** o1 is OpenAI's December 2024 reasoning model, the successor to the o1-preview. It scores 97.3% on GSM8K and 94.8% on the MATH benchmark, demonstrating strong quantitative reasoning. Priced at $15 input / $60 output per 1M tokens with a 200K-token context window. Succeeded by o3 (April 2025), which delivers substantially better performance at lower cost. ## At a Glance **Where o1 leads** - 97.3% GSM8K — top-tier grade-school mathematics - 94.8% MATH — strong competition math performance - Pioneer of the o-series chain-of-thought paradigm - Multimodal vision support - 200K-token context window **Where it lags** - Superseded by o3 (April 2025), which is stronger across all benchmarks at lower cost ($10/$40 vs $15/$60) - 7.96% Humanity's Last Exam — below subsequent o-series models - Highest per-token price in the o-series - Closed API model with no open weights **Best for:** Teams with existing o1 integrations; historical reference; workflows where the specific o1 behaviour has been validated in production. ## What o1 Is o1 (December 2024) was OpenAI's first fully released o-series model, following the earlier o1-preview (September 2024). It introduced chain-of-thought reasoning to the mainstream OpenAI API: the model generates internal reasoning tokens before producing a final answer, improving accuracy on hard math, science, and coding tasks at the cost of higher latency and token usage. At its December 2024 launch, o1 was state-of-the-art on competitive mathematics benchmarks (97.3% GSM8K, 94.8% MATH) and represented a step-change over GPT-4o for tasks requiring sustained multi-step reasoning. It established the o-series as OpenAI's primary line for hard reasoning tasks. o1 was superseded by o3 in April 2025, which delivers materially stronger results across all benchmarks at a lower price ($10/$40 vs o1's $15/$60). For new integrations, o3 or o4-mini is the recommended choice. o1 remains available via the API for teams with existing validated deployments. ## Specifications | Field | Value | |---|---| | Organization | OpenAI | | Parameters | Undisclosed | | Context window | 200,000 tokens | | Max output | 100,000 tokens | | Architecture | Chain-of-thought reasoning (o-series) | | License | Proprietary (API only) | | Release date | December 17, 2024 | | Knowledge cutoff | October 2023 | | Modality | Text + Vision (multimodal) | ## Pricing | | Input (per 1M tokens) | Output (per 1M tokens) | |---|---|---| | OpenAI API | $15.00 | $60.00 | Reasoning tokens are billed as output tokens. Pricing per [OpenAI pricing page](https://openai.com/api/pricing/). Note: o3 provides better performance at $10/$40 per 1M. ## Context Window o1 has a **200,000-token context window** — roughly **150 pages** of text in a single request, matching o3 and o4-mini. ## Public Benchmark Scores | Benchmark | Score | Source | Date | |---|---|---|---| | [GSM8K](/benchmarks/gsm8k) | 97.3% | Benchgen evaluation | 2025-07 | | [MATH](/benchmarks/math) | 94.8% | Benchgen evaluation | 2025-07 | | [Humanity's Last Exam](/benchmarks/humanitys-last-exam) | 7.96% | Benchgen evaluation | 2025-07 | *Scores from Benchgen evaluations. See the source for methodology.* ## o1 vs Alternatives | Model | Context | GSM8K | MATH | Price (in/out per 1M) | |---|---|---|---|---| | **o1** | 200K | 97.3% | 94.8% | $15 / $60 | | [o3](/models/openai/o3) | 200K | — | — | $10 / $40 | | [o4-mini](/models/openai/o4-mini) | 200K | — | — | $1.10 / $4.40 | | [DeepSeek R1](/models/deepseek/r1) | 128K | — | — | ~$0.55 / $2.19 | o3 and o4-mini both outperform o1 on modern benchmarks at lower cost. For new deployments, o4-mini is the recommended default; for maximum reasoning capability, o3 is the better choice. ## Use o1 via API ```python from openai import OpenAI client = OpenAI(api_key="YOUR_API_KEY") response = client.chat.completions.create( model="o1", messages=[{"role": "user", "content": "Prove that the square root of 2 is irrational..."}], ) print(response.choices[0].message.content) ``` ## Related Models & Benchmarks - Compare: [o3](/models/openai/o3) · [o4-mini](/models/openai/o4-mini) · [o1 Pro](/models/openai/o1-pro) - Benchmarks: [GSM8K](/benchmarks/gsm8k) · [MATH](/benchmarks/math) · [Humanity's Last Exam](/benchmarks/humanitys-last-exam) ## Frequently Asked Questions

What is o1?

o1 is OpenAI's December 2024 reasoning model, the first fully released member of the o-series. It uses chain-of-thought reasoning to achieve strong performance on math and science benchmarks. It has been superseded by o3 (April 2025), which is stronger and cheaper.

What is o1's context window?

o1 supports a 200,000-token context window with up to 100,000 output tokens.

How much does o1 cost?

o1 is priced at $15.00 per 1M input tokens and $60.00 per 1M output tokens. o3 provides better performance at $10/$40 per 1M and is the recommended alternative.

Is o1 open source?

No. o1 is a proprietary, API-only model. There are no open weights.

Should I use o1 or o3?

For new integrations, o3 is the better choice: it outperforms o1 on all benchmarks and costs $10/$40 per 1M (vs o1's $15/$60). o1 is appropriate only if you have an existing production deployment validated on o1-specific behaviour.

--- *Specs and scores sourced from OpenAI's official o1 announcement (December 17, 2024) and Benchgen evaluations. Pricing cited to the OpenAI pricing page. Last updated 2026-07-23.*

o1 — Evaluations

Evaluations