OpenAI · The Reasoning-Model Era
OpenAI o3
OpenAI's peak reasoning model before GPT-5. AIME, ARC-AGI, and SWE-bench records at release.
By C.W. Jameson · Published 19 May 2026 · Last reviewed 19 May 2026
o3 was the clearest demonstration that chain-of-thought at scale produces qualitatively different outputs from next-token prediction. It solved problems that had defeated GPT-4 class models for years. The cost per query at high effort was prohibitive; the capability was not.
Field signature
Reasoning tokens billed separately. Full answer often shorter than the thinking trace.
Specifications
| Released | 2025-04 |
|---|---|
| Context window | 200,000 tokens |
| Pricing | $10 / $40 per million tokens (varies by effort) |
| Modalities | text · image |
| License | Commercial API only |
| Era | The Reasoning-Model Era |
Strengths
- Competition math
- Long-horizon problem decomposition
- Code audit
Weaknesses
- Cost at high reasoning effort
- Latency
Authentication markers
The fingerprints by which OpenAI o3 can be identified from its output alone.
| Tell | Meaning |
|---|---|
| Reasoning trace visible in extended output. | o-series model. |
Notable works
- ARC-AGI 87.5% at high compute — first to approach human-level on that benchmark
Market position
$10-$40 per million tokens
Partner offer
OpenAI's API surface remains the broadest commercial offering.
Try OpenAI →Affiliate link — see disclosure.
Primary sources
- [1] OpenAI: o3
From the Almanac shop
The Operator's Compendium
Every agent harness, every routing pattern, every cost trick. 90-page PDF.
$29 — Coming soon