o3
OpenAI · o-series
Still-available OpenAI reasoning model for difficult analysis, retained as a legacy reference after GPT-5 became the default recommendation.
Overview
Freshness note: Model capabilities, limits, and pricing can change quickly. This profile is a point-in-time snapshot last verified on March 27, 2026.
o3 is still available in OpenAI’s API, but it is no longer the default direction OpenAI recommends for most new work. OpenAI’s current model guidance positions GPT-5 and GPT-5 mini as the main starting points, while o3 now reads more like a compatibility and specialist reasoning route for teams that still prefer the older o-series behavior profile.
That makes o3 worth keeping in the index, but not worth treating as a current default. The model is best understood as a legacy reasoning option that remains useful for some hard-analysis workloads.
Capabilities
o3 is strongest on difficult multi-step reasoning, constraint-heavy planning, quantitative analysis, and ambiguity resolution. It is still a credible route for high-stakes technical or operational reasoning when teams already know they prefer the o-series style over the newer GPT-5 family defaults.
Technical Details
OpenAI’s current model docs still list o3 with a 200K context window, 100K max output tokens, and support for image input. The reasoning focus remains the point: this is not a throughput-optimized generalist, but a model built for longer analytical work.
The larger strategic detail is model-line direction. OpenAI now treats GPT-5 as the main public recommendation, so o3 is better framed as an older specialized option than as a frontline default.
Pricing & Access
Published standard API pricing (per 1M tokens):
- Input: $2.00
- Output: $8.00
OpenAI’s Batch API pricing effectively halves that to:
- Input: $1.00
- Output: $4.00
Access is through current OpenAI API model endpoints for accounts with the relevant model availability.
Best Use Cases
Use o3 when you specifically want a still-supported OpenAI reasoning model for architecture tradeoff analysis, policy interpretation, migration planning, incident root-cause reasoning, or similar tasks where careful stepwise analysis matters more than raw cost efficiency.
For new OpenAI builds without a compatibility reason, GPT-5 or GPT-5 mini are usually the better starting points.
Comparisons
- GPT-5 (OpenAI): Current default direction for most new OpenAI work, with broader product momentum than o3.
- o4-mini (OpenAI): Cheaper reasoning route, but with less headroom for harder tasks.
- Claude Opus 4.6 (Anthropic): Strong alternative when teams want a current premium reasoning model outside the OpenAI stack.