OpenAI's powerful reasoning model that pushes the frontier across coding, math, science, and visual perception. Excels in complex queries requiring multi-faceted analysis. Succeeded by GPT-5.
OpenAI o3 is the flagship of the o-series reasoning generation, released on April 16, 2025 alongside o4-mini. It is trained to spend more compute thinking before responding, and at launch it set new state-of-the-art results on Codeforces (Elo 2727), SWE-bench Verified (71.7%), and MMMU. On TheRouter, openai/o3 is exposed through the standard /v1/chat/completions endpoint at $2 per 1M input tokens and $8 per 1M output tokens, with a $0.50 cached-input rate.
OpenAI's own docs label o3 as "succeeded by GPT-5," but the model is still active and not on any deprecation list — it remains the right pick when you want a reasoning model with a known, mid-2024 knowledge cutoff, native vision input, and a 200K context with 100K max output. Use o3 when GPT-5 family pricing or behaviour is a poor fit and you need pre-GPT-5 stability; reach for gpt-5 / gpt-5-pro when you can absorb the price-per-quality jump.
| Type | Rate |
|---|---|
| Input | $2.16 / 1M tokens |
| Output | $8.64 / 1M tokens |
| Release date | 2025-04-16openai.com ↗ | verified |
| Snapshot | o3-2025-04-16developers.openai.com ↗ | verified |
| Knowledge cutoff | June 1, 2024developers.openai.com ↗ | verified |
| Context window | 200,000 tokensdevelopers.openai.com ↗ | verified |
| Max output tokens | 100,000 tokensdevelopers.openai.com ↗ | verified |
| Reasoning tokens | Supported (reasoning_effort: low | medium | high)developers.openai.com ↗ | verified |
| Status | Active. Marked as "succeeded by GPT-5" in OpenAI docs but not on any deprecation list as of 2026-05-31.developers.openai.com/deprecations ↗ | verified |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
Codeforces Approximately top 0.2% / ~#175 globally on Codeforces at the time of evaluation. | Elo 2727Elo | openai.com ↗ | |
SWE-bench Verified Reported by OpenAI without a model-specific custom scaffold; n=477 verified subset. | 71.7%% | openai.com ↗ | |
AIME 2024 OpenAI announcement post (Dec 2024 frontier release). Pass@1 with tool access (Python). | 96.7% (with tools)% | openai.com ↗ | |
AIME 2025 Reported by OpenAI: 98.4% pass@1 / 100% consensus@8 with Python interpreter. | 98.4% pass@1 (with tools)% | openai.com ↗ | |
GPQA Diamond Third-party comparison summary. PhD-level science multiple choice. | 83.3%% | datacamp.com ↗ | |
MMMU OpenAI claimed SOTA on MMMU at the o3/o4-mini launch; numerical figure not given in the launch post. | SOTA at launch | openai.com ↗ |
Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.
curl https://api.therouter.ai/v1/chat/completions -H "Content-Type: application/json" -H "Authorization: Bearer $THE_ROUTER_API_KEY" -d '{
"model": "openai/o3",
"messages": [
{"role": "user", "content": "Summarize the key points from this input."}
]
}'Standard OpenAI-compatible chat. o3 is available via /v1/chat/completions on TheRouter — same endpoint, same SDK as gpt-4o or gpt-5.
curl https://api.therouter.ai/v1/chat/completions \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "openai/o3",
"messages": [
{"role": "user", "content": "Prove that the sum of two odd numbers is even."}
]
}'OpenAI shipped o3-pro on June 10, 2025 — a higher-effort version of o3 that thinks longer for more reliable answers. For TheRouter callers this means o3 (this page) stays the cost-balanced reasoning option, while openai/o3-pro is the right pick when you want maximum reliability per query.
OpenAI shipped o3 and o4-mini on April 16, 2025, framing them as the first o-series models that can agentically use tools — web, Python, file analysis, image generation. Routing-layer takeaway: o3 is the deeper, slower, more expensive sibling; o4-mini is the high-throughput option for the same shape of workload.
No. As of 2026-05-31 OpenAI's deprecations page does NOT list o3 (only o3-mini and o3-deep-research are scheduled for shutdown). The o3 model page does, however, label it "succeeded by GPT-5," which is OpenAI's way of telling you the long-term migration target.
200,000 tokens of context and 100,000 max output tokens, per OpenAI's official model docs.
Yes — text and image input, text-only output. o3 was the first o-series reasoning model with first-class vision input, and OpenAI specifically called out strong performance on charts, diagrams, and images at launch.
o3 is priced at $2 input / $8 output per 1M tokens with $0.50 cached input, materially cheaper than the GPT-5 reasoning tiers (gpt-5-pro is $18.75/$150 on TheRouter). For workloads where o3-grade reasoning is enough, staying on o3 is the cost-rational call until OpenAI publishes a deprecation date.
| source | URL | retrieved | |
|---|---|---|---|
| Release date | openai.com ↗ | 2026-05-31 | verified |
| Snapshot | developers.openai.com ↗ | 2026-05-31 | verified |
| Knowledge cutoff | developers.openai.com ↗ | 2026-05-31 | verified |
| Context window | developers.openai.com ↗ | 2026-05-31 | verified |
| Max output tokens | developers.openai.com ↗ | 2026-05-31 | verified |
| Reasoning tokens | developers.openai.com ↗ | 2026-05-31 | verified |
| Status | developers.openai.com/deprecations ↗ | 2026-05-31 | verified |
| Codeforces | openai.com ↗ | 2026-05-31 | verified |
| SWE-bench Verified | openai.com ↗ | 2026-05-31 | verified |
| AIME 2024 | openai.com ↗ | 2026-05-31 | verified |
| AIME 2025 | openai.com ↗ | 2026-05-31 | verified |
| GPQA Diamond | datacamp.com ↗ | 2026-05-31 | to verify |
| MMMU | openai.com ↗ | 2026-05-31 | to verify |
| o3-pro available to API and ChatGPT Pro | OpenAI Help Center ↗ | 2026-05-31 | verified |
| OpenAI o3 and o4-mini launch | openai.com ↗ | 2026-05-31 | verified |
| Is o3 deprecated yet? | developers.openai.com ↗ | 2026-05-31 | to verify |
| What is o3's context window and max output? | developers.openai.com ↗ | 2026-05-31 | to verify |
| Does o3 accept image input? | developers.openai.com ↗ | 2026-05-31 | to verify |
| How does o3 compare to GPT-5 on price? | developers.openai.com ↗ | 2026-05-31 | to verify |