GPT-o3
OpenAI GPT-o3: a full-scale reasoning model featuring reinforcement learning scaling, RL-trained tool use, and image reasoning inside the chain of thought.
Model Overview
Capabilities, design details, and architectural traits
GPT-o3 - Reinforcement Learning Scaling Applied to Reasoning
GPT-o3 is OpenAI's full-scale o-series reasoning model, built around a documented finding: large-scale reinforcement learning exhibits the same "more compute = better performance" scaling trend observed in GPT-series pretraining. OpenAI pushed an additional order of magnitude in both training compute and inference-time reasoning for GPT-o3, and performance continued to improve.
| Trait | Detail |
|---|---|
| Images in the chain of thought | GPT-o3 integrates images directly into its reasoning process - it thinks with images, not just about them; it can process blurry, reversed, or low-quality images and manipulate them (rotate, zoom, transform) mid-reasoning |
| RL-trained tool use | Trained via reinforcement learning to reason about when and how to use tools, not just how - tool deployment is based on desired outcomes |
| Reasoning effort is configurable | The reasoning_effort parameter controls how many reasoning tokens are allocated per request |
| API availability | Available via both Chat Completions and Responses API; the Responses API additionally supports reasoning summaries and preserving reasoning tokens across function calls |
| Fine-tuning | Not supported |
Designed for Multi-Step Problems Across Modalities
OpenAI's official positioning is to use GPT-o3 for multi-step problems that involve analysis across text, code, and images. It is suited to technical writing and instruction-following in addition to math, science, and coding - a broader documented scope than earlier o-series models.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | OpenAI | Anthropic | Anthropic |
| Release Date | April 16, 2025 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | Jun 2024 | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 200K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | $2 Best Input Pricing | $4 | $2 Best Input Pricing |
| Output Pricing | $8 Best Output Pricing | $20 | $10 |
| Modalities | |||
| Inputs | imagetextfile | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 20.2 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | - | - | - |
| Agentic Index | - | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from OpenAI
Other models by OpenAI
Top AI Models
Leading alternatives by intelligence score
