News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
O3

O3

Reasoning model from OpenAI designed for step-by-step logical tasks.

OpenAI released o3 in April 2025 as a successor to o1. It uses reinforcement learning to plan answers before generating them. This approach improves performance on complex science and coding tasks.

Share on WhatsApp

  • Announced in December 2024 to avoid O2 trademark conflict.
  • Uses private chain of thought for complex reasoning tasks.
  • Achieved 87.7% on GPQA Diamond expert science benchmark.
  • Scored 71.7% on SWE-bench Verified software engineering test.
  • Retirement announced in May 2026 for August 2026 removal.

What it is

OpenAI o3 is a reasoning language model built by OpenAI. It serves as the next generation after o1 for the ChatGPT platform. The system focuses on complex logic by spending extra time thinking through problems. It uses a private chain of thought to plan steps before answering. This method increases latency but boosts accuracy on hard tasks. A smaller version called o3-mini arrived earlier in January 2025. The main o3 model launched in April 2025 alongside o4-mini.

How it came about

OpenAI announced the project in December 2024. They chose the name o3 to avoid trademark issues with the O2 mobile brand. Researchers received early access until January 2025. The team released o3-mini first, offering it to free and paid users. Updates in February improved transparency and rate limits. The full o3 model followed in April. Later that year, OpenAI introduced o3-pro for high-stakes reliability. By May 2026, the company planned to retire the original o3 model in August.

Inside the model

The model relies on reinforcement learning to teach itself how to reason. It generates a private chain of thought before producing final output. This internal planning allows the system to solve multi-step problems more effectively. The design requires significantly more computing power than standard models. Users can adjust effort levels on the mini variant. Higher compute settings are reserved for subscribers. The architecture remains a generative pre-trained transformer focused on text processing.

What it can do

o3 outperforms its predecessor o1 across several benchmarks. It scored 87.7% on the GPQA Diamond test for expert science questions. On the SWE-bench Verified software engineering challenge, it achieved 71.7%. The model reached an Elo rating of 2727 on Codeforces compared to 1891 for o1. It also demonstrated three times the accuracy on the ARC-AGI benchmark. These results highlight strong capabilities in mathematics, coding, and scientific reasoning.

How to use it

Users access o3 through the ChatGPT application and API services. The o3-mini variant became available to all users in January 2025. Paid subscribers gained access to higher compute options. The full o3 model launched via API and ChatGPT in April 2025. OpenAI Deep Research utilizes a specific version of the model for report generation. Access terms vary by subscription tier, with Pro users receiving unlimited usage for certain variants.

Safety and testing

OpenAI invited security researchers to review the models before public release. The company emphasized safety during the development phase. Updates in February aimed to increase transparency in the reasoning process. This change allowed users to see parts of the thought process. The organization continues to monitor model behavior through these research partnerships. No major safety incidents or bans were reported in the provided material.

Reception and debate

The model received attention for its improved reasoning capabilities. Rivals like DeepSeek R1 pressured OpenAI to enhance transparency. The company responded by exposing more of the internal thought steps. Performance gains on technical benchmarks drew praise from developers. The retirement announcement in 2026 marked a transition to newer architectures. Users adapted to the changing availability of different model tiers.

Why it matters

o3 set new standards for reasoning tasks in large language models. Its success on benchmarks influenced the direction of subsequent releases. The introduction of o3-pro expanded options for reliability-critical applications. The retirement of o3 paved the way for o4-mini and GPT-5. The model demonstrated the value of extended deliberation in AI systems. Its architecture informed future developments in the OpenAI product line.

Questions people ask about O3

What is the main purpose of OpenAI o3?

OpenAI o3 is a reasoning model designed for step-by-step logical tasks. It uses reinforcement learning to plan answers internally. This approach improves performance on complex coding and science problems.

How does o3 differ from o1?

o3 replaces o1 as the successor model. It demonstrates significantly better performance on benchmarks. For example, it scored 71.7% on SWE-bench versus 48.9% for o1. It also reaches higher Elo ratings on competitive coding platforms.

When was o3 released?

OpenAI announced o3 in December 2024. The smaller o3-mini variant launched on January 31, 2025. The main o3 model and o4-mini followed on April 16, 2025.

Why was the name o3 chosen?

The name o3 was selected to avoid trademark conflicts with the mobile carrier brand O2. This naming decision allowed OpenAI to continue its sequential numbering without legal issues.

Good reading and sources

Where this page comes from

The facts on this page come from the Wikipedia article "OpenAI o3", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryLLM rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.

It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-15.

The weekly digest

Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.