# OpenAI o3-mini

> Source: https://aiwiki.ai/wiki/o3_mini
> Updated: 2026-10-28
> Fact-checked: 2026-11-05
> Categories: Large Language Models, OpenAI, Reasoning Models
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "OpenAI o3-mini." aiwiki.ai, 28 Oct 2026. https://aiwiki.ai/wiki/o3_mini
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**OpenAI o3-mini** is a reasoning-focused [large language model](https://aiwiki.ai/wiki/large_language_model) released by [OpenAI](https://aiwiki.ai/wiki/openai) on January 31, 2025, the second commercial member of the o-series after [OpenAI o1](https://aiwiki.ai/wiki/o1) and a smaller, cheaper, faster sibling to the full o3 model that was previewed alongside it.[^1][^2] It was the first OpenAI [reasoning model](https://aiwiki.ai/wiki/reasoning_models) made available to free [ChatGPT](https://aiwiki.ai/wiki/chatgpt) users, and the first to support [function calling](https://aiwiki.ai/wiki/function_calling), [structured output](https://aiwiki.ai/wiki/structured_output)s, and developer messages from launch; it also introduced a user-selectable `reasoning_effort` parameter with low, medium, and high settings.[^2][^24] OpenAI positioned the model as STEM-focused, calling it "a specialized alternative" to o1 "for technical domains requiring precision and speed" with particular strength in science, mathematics, and coding.[^1][^24] At launch its API price was $1.10 per million input tokens and $4.40 per million output tokens, roughly 63% cheaper per input token than o1-mini and competitive with [DeepSeek-R1](https://aiwiki.ai/wiki/deepseek_r1), which had been released eleven days earlier on January 20, 2025.[^2][^3][^24] The charts in OpenAI's launch post show the high-effort setting at 87.3% on [AIME 2024](https://aiwiki.ai/wiki/aime_2024), 79.7% on [GPQA Diamond](https://aiwiki.ai/wiki/gpqa_diamond) and a Codeforces rating of 2130 Elo; launch-day coverage and OpenAI's April 2025 comparison used lower figures for some of these, including 77.0% on GPQA Diamond and 2073 Elo.[^2][^4][^24] It was eventually superseded by [o4-mini](https://aiwiki.ai/wiki/o4_mini) on April 16, 2025, and OpenAI shut it down in the [OpenAI API](https://aiwiki.ai/wiki/openai_api) on October 23, 2026, after announcing its deprecation on April 22, 2026.[^4][^5][^26]

<figure class="w-56">
<img src="/images/wd-2026-10-28-openai-logo-2a509640a546-tile.png" alt="OpenAI logo" width="480" height="178" />
<figcaption>OpenAI logo. Source: <a href="https://commons.wikimedia.org/wiki/File:OpenAI_logo_2025_(wordmark).svg">Wikimedia Commons</a>, <a href="https://commons.wikimedia.org/wiki/File:OpenAI_logo_2025_(wordmark).svg#Licensing">Public domain</a>.</figcaption>
</figure>

## Infobox

| Field | Value |
| --- | --- |
| Developer | [OpenAI](https://aiwiki.ai/wiki/openai) |
| Announced | December 20, 2024 (preview)[^6] |
| General availability | January 31, 2025[^2] |
| API model ID | `o3-mini` and `o3-mini-2025-01-31`[^7] |
| [Context window](https://aiwiki.ai/wiki/context_window) | 200,000 tokens (max 100,000 output)[^7] |
| Knowledge cutoff | October 1, 2023[^7] |
| Modalities at launch | Text only (image input added in [ChatGPT](https://aiwiki.ai/wiki/chatgpt) on February 13, 2025)[^8] |
| Reasoning effort levels | low, medium (default), high[^2][^7] |
| API price (input / cached / output) | $1.10 / $0.55 / $4.40 per 1M tokens[^7][^13] |
| Replaced by | [o4-mini](https://aiwiki.ai/wiki/o4_mini) (April 16, 2025)[^4] |

<figure class="embedvideo">
<iframe src="https://www.youtube-nocookie.com/embed/SKBG1sqdyIU" title="OpenAI o3 and o3-mini: 12 Days of OpenAI: Day 12"></iframe>
<figcaption>"OpenAI o3 and o3-mini: 12 Days of OpenAI: Day 12", official video from OpenAI, published December 20, 2024. <a href="https://www.youtube.com/watch?v=SKBG1sqdyIU">Watch on YouTube</a>.</figcaption>
</figure>

## When was OpenAI o3-mini announced and released?

OpenAI introduced the o3 family during the final day of its "12 Days of OpenAI" livestream series on December 20, 2024. CEO [Sam Altman](https://aiwiki.ai/wiki/sam_altman) and researchers [Mark Chen](https://aiwiki.ai/wiki/mark_chen), Hongyu Ren, and others demonstrated the full o3 model and announced a sibling called o3-mini, billed as a smaller, faster variant suited to coding and STEM tasks where the full model would be excessive.[^6] OpenAI skipped the "o2" name to avoid a trademark conflict with the British telecommunications brand O2, a detail Altman discussed during the stream.[^6] The December event opened a safety testing window: external red teamers and safety researchers could apply for early access to both o3 and o3-mini, with deployment dates contingent on the outcome of those evaluations.[^6]

On January 31, 2025, OpenAI shipped o3-mini in both ChatGPT and the API. The launch followed by eleven days the public release of DeepSeek-R1 on January 20, 2025, which had upended pricing expectations for reasoning models by offering comparable benchmark performance at a fraction of the cost.[^3][^9] OpenAI's announcement page described o3-mini as "the newest, most cost-efficient model in our reasoning series" and emphasised that it was production-ready out of the gate thanks to function calling, structured outputs, and developer messages.[^2][^24] Pro subscribers received unlimited access to both o3-mini and a higher-thinking variant called o3-mini-high; Plus and Team subscribers received 150 messages per day, triple the cap that had previously applied to o1-mini.[^2] Enterprise and Education customers received access "in a week," and free ChatGPT users could trigger the model by selecting a "Reason" button in the message composer, the first time a reasoning model had reached the free tier.[^2][^25]

On February 12 and 13, 2025, OpenAI quietly enabled image upload in ChatGPT for o3-mini accounts, though vision inputs remained unsupported in the public API at that time.[^8] In February 2025 OpenAI raised the o3-mini-high allowance for Plus subscribers to 50 messages per day, which the cited write-up describes as seven times the earlier limit.[^10]

The full o3 reasoning model and a new mini sibling, o4-mini, launched on April 16, 2025, and An OpenAI employee described the change in ChatGPT as a straight one-to-one upgrade in which o4-mini replaced o3-mini in the model picker.[^4][^5] OpenAI kept o3-mini available in the API to preserve stable behaviour for production users that had built on it.[^5] On June 20, 2025, [GitHub](https://aiwiki.ai/wiki/github) announced the deprecation of o1, [GPT-4.5](https://aiwiki.ai/wiki/gpt-4.5), o3-mini, and [GPT-4o](https://aiwiki.ai/wiki/gpt_4o) in GitHub Copilot, with o3-mini originally scheduled for retirement there on July 18, 2025; GitHub later marked that o3-mini deprecation as postponed, and OpenAI continued listing `o3-mini` in its API catalogue through the second half of 2025.[^11]

## What can OpenAI o3-mini do?

### What does the reasoning_effort parameter do?

The most prominent design decision in o3-mini is the introduction of an explicit `reasoning_effort` parameter, accepting the string values `low`, `medium`, and `high`. The parameter trades latency for accuracy by allowing the model to allocate more or fewer hidden reasoning tokens before producing the final answer.[^2][^7] OpenAI's launch post describes the choice as a way to "think harder when tackling complex challenges or prioritize speed when latency is a concern."[^1][^24] Medium effort is the default in both ChatGPT and the API, while developers can override the setting per API call.[^2][^7] At high effort, OpenAI reported that o3-mini outperformed o1 on competition math and coding, while describing its GPQA Diamond science score as comparable with o1's.[^2][^24] At low effort it roughly matched o1-mini, and at medium effort it roughly matched the full o1.[^24]

The three effort tiers showed a roughly monotonic improvement on benchmark scores. On the 2024 American Invitational Mathematics Examination, OpenAI's launch-post chart shows pass@1 accuracy of 60.0% at low effort, 79.6% at medium, and 87.3% at high effort with no Python [tool use](https://aiwiki.ai/wiki/tool_use); TechCrunch's launch-day report, however, said o3-mini-high beat o1's 83.3% by just 0.3 percentage points.[^2][^24] Codeforces ratings in the same post rise from 1831 (low) and 2036 (medium) to 2130 (high); OpenAI's April 2025 o3 and o4-mini announcement lists o3-mini at 2073.[^24][^4]

In ChatGPT, OpenAI exposed only two tiers to consumers: regular o3-mini (medium effort) and a separately badged o3-mini-high variant that ran at high effort and consumed a smaller daily quota.[^2]

### Does o3-mini support function calling and structured outputs?

Yes. Unlike its predecessor o1-mini, o3-mini natively supports the function-calling API that OpenAI shipped for [GPT-4](https://aiwiki.ai/wiki/gpt-4) in 2023, the Structured Outputs feature added for GPT-4o in August 2024, and the new `developer` role that OpenAI introduced for the o-series in late 2024.[^2][^24] The launch post called these "highly requested developer features" and framed o3-mini as OpenAI's "first small reasoning model that supports" them, "making it production-ready out of the gate," in contrast to o1 and o1-mini, both of which had launched without function calling or strict JSON-schema output enforcement.[^2][^24]

Structured Outputs in o3-mini lets developers pass a JSON Schema in the API request and receive a response whose structure is guaranteed to validate against the schema, removing the need for retry-on-parse-error loops common in LLM pipelines.[^2][^24] Function calling allows the model to emit machine-readable tool invocations against a developer-supplied list of available functions, enabling tool use, [agentic workflow](https://aiwiki.ai/wiki/agentic_workflow)s, and [retrieval-augmented generation](https://aiwiki.ai/wiki/retrieval_augmented_generation).[^2] Developer messages occupy a system-message-like slot in the request payload but are designed for the o-series, where models trained to reason internally treat them as high-trust instructions distinct from end-user content.[^2][^24]

### Which modalities does o3-mini support?

At the January 31, 2025 launch, o3-mini accepted only text input and produced only text output, with no vision or audio support.[^2][^7] OpenAI's launch documentation acknowledged that the model "does not support vision capabilities, so developers should continue using OpenAI o1 for visual reasoning tasks."[^2][^24] Two weeks later, on February 12 to 13, 2025, image upload became available for o3-mini chats in the ChatGPT web app, with OpenAI confirming the rollout publicly the following day; the API still rejected `image_url` payloads against the o3-mini endpoint at that time.[^8]

### Does o3-mini support web search?

OpenAI shipped a prototype web-search integration alongside o3-mini in ChatGPT. The model can fetch up-to-date information and surface citations to source pages, though OpenAI labelled the feature as an early prototype while it worked to integrate search across the reasoning-model family.[^2][^24] Full agentic web-search and tool use, including Python execution and image generation from within the chain of thought, was reserved for the later o4-mini and o3 launches in April 2025.[^4]

### What are the context window and output limits?

The o3-mini model has a 200,000-token [context window](https://aiwiki.ai/wiki/context_window) with a maximum output of 100,000 tokens, including the hidden reasoning tokens consumed at higher effort settings.[^7] The knowledge cutoff is October 1, 2023, the same as o1.[^7]

## How does OpenAI o3-mini perform on benchmarks?

OpenAI's launch post for o3-mini included evaluation results across mathematics, science, coding, and tool-use benchmarks. Unless stated otherwise the figures below are pass@1 with no external tools and follow OpenAI's reported numbers for the three effort levels.[^2]

### Mathematics

On the 2024 American Invitational Mathematics Examination competition problems, OpenAI's launch-post chart shows 60.0% at low effort, 79.6% at medium, and 87.3% at high effort, and OpenAI's April 2025 o3 and o4-mini announcement also lists o3-mini at 87.3%.[^24][^4] TechCrunch's launch-day report described a smaller margin, saying o3-mini-high beat o1 by just 0.3 percentage points.[^2]

On OpenAI's [MATH](https://aiwiki.ai/wiki/math) evaluation, o3-mini scored 95.8% at low effort, 97.3% at medium and 97.9% at high effort, compared with 90.0% for o1-mini.[^24] On the [FrontierMath](https://aiwiki.ai/wiki/frontiermath) research-math benchmark introduced by [Epoch AI](https://aiwiki.ai/wiki/epoch_ai) in late 2024, OpenAI reported that o3-mini at high effort with access to a Python tool solved over 32% of problems on the first attempt, including more than 28% of the "T3" hardest problems.[^2] For context, when OpenAI previewed o3 in December 2024, TechCrunch reported that no other model exceeded 2% on FrontierMath.[^6]

### Science

On GPQA Diamond, the 198-question subset of GPQA covering graduate-level physics, chemistry, and biology questions, OpenAI's launch-post chart shows o3-mini at 70.6% at low effort, 76.8% at medium, and 79.7% at high effort, against 78.0% for o1 and 60.0% for o1-mini.[^24] TechCrunch's launch-day report said o3-mini did not surpass o1's score even at high effort, and OpenAI's April 2025 comparison lists o3-mini at 77.0% against o1's 78.0%.[^2][^4]

### Coding

On the Codeforces competitive-programming Elo ladder, OpenAI's launch-post chart assigned ratings of 1831, 2036, and 2130 to o3-mini at low, medium, and high effort respectively, against 1891 for o1; OpenAI's April 2025 comparison lists o3-mini at 2073.[^24][^4] Both figures sit well below the 2727 rating OpenAI reported for the full o3 model in December 2024.[^6]

On [SWE-bench Verified](https://aiwiki.ai/wiki/swe-bench_verified), the human-curated 500-problem subset of [SWE-bench](https://aiwiki.ai/wiki/swe_bench) drawn from real GitHub pull requests, o3-mini scored 40.8% at low effort, 42.9% at medium and 49.3% at high effort on OpenAI's 477-task subset.[^24] The high-effort 49.3% figure compared favourably to o1's 48.9% on the same evaluation.[^2] On the [LiveBench](https://aiwiki.ai/wiki/livebench) coding category, OpenAI reported that o3-mini surpassed o1-high even at medium reasoning effort, with a wider lead at high effort.[^24]

### Knowledge and general capabilities

A notable trade-off in o3-mini was its weaker performance on broad knowledge benchmarks. On the MMLU multitask language understanding benchmark, o3-mini at high effort scored 86.9%, several points below o1's reported 92.3%, reflecting the model's relative narrowing of training toward STEM reasoning rather than encyclopedic recall.[^13]

### Comparative scoreboard

| Benchmark | o3-mini (high) | [o1](https://aiwiki.ai/wiki/o1) | [DeepSeek-R1](https://aiwiki.ai/wiki/deepseek_r1) | Source |
| --- | --- | --- | --- | --- |
| [AIME 2024](https://aiwiki.ai/wiki/aime_2024) (pass@1) | 87.3% | 83.3% | 79.8% | [^9][^24] |
| [GPQA Diamond](https://aiwiki.ai/wiki/gpqa_diamond) | 79.7% (77.0% in OpenAI's April 2025 comparison) | 78.0% | 71.5% | [^4][^9][^24] |
| MATH-500 | 97.9% (OpenAI's MATH eval) | 96.4% | 97.3% | [^9][^24] |
| Codeforces Elo | 2130 (2073 in OpenAI's April 2025 comparison) | 1891 | 2029 | [^4][^9][^24] |
| [SWE-bench Verified](https://aiwiki.ai/wiki/swe-bench_verified) | 49.3% | 48.9% | 49.2% | [^2][^9] |
| [FrontierMath](https://aiwiki.ai/wiki/frontiermath) (with Python) | >32% | not reported | not reported | [^2] |
| MMLU | 86.9% | 92.3% | 90.8% | [^13][^9] |

### How does o3-mini compare on human preference and speed?

OpenAI also reported the results of an "external expert tester" study in which evaluators preferred o3-mini responses to o1-mini responses 56% of the time and judged that o3-mini reduced "major mistakes" on real-world hard questions by 39%.[^2][^24] The same study found average response latency was 24% faster than o1-mini, at 7.7 seconds versus 10.16 seconds, with an average time-to-first-token about 2500 milliseconds faster than o1-mini.[^2][^24]

## How was OpenAI o3-mini trained?

OpenAI has not published a detailed architectural report for o3-mini. The o3-mini system card describes the model as a member of the o-series trained with large-scale [reinforcement learning](https://aiwiki.ai/wiki/reinforcement_learning) to produce long internal [chain-of-thought](https://aiwiki.ai/wiki/chain_of_thought) traces before responding.[^15] The system card states that "OpenAI o3-mini is the latest model in this series" and explains that, through reinforcement-learning training, models in the family "learn to refine their thinking process, try different strategies, and recognize their mistakes."[^15] Like the rest of the o-series, o3-mini consumes a configurable number of hidden reasoning tokens during inference, and OpenAI bills those reasoning tokens at the output rate even though they are not visible in the final response.[^7]

Beyond reasoning training, OpenAI applied a safety-training procedure called deliberative alignment to o3-mini. Introduced in a paper published alongside the December 2024 o3 preview, deliberative alignment teaches a model to explicitly reason through OpenAI's safety policy text before producing an answer, rather than relying only on examples of compliant and refusal behaviour.[^16] OpenAI's deliberative-alignment paper lists o1-preview, o1, and o3-mini as the production models trained with the technique, and the o3-mini system card credits deliberative alignment for improvements in jailbreak resistance and policy compliance.[^16][^15]

## How safe is OpenAI o3-mini?

The o3-mini system card, dated January 31, 2025 and revised on February 10, 2025, evaluates o3-mini against OpenAI's [Preparedness Framework](https://aiwiki.ai/wiki/preparedness_framework) in four risk categories: Cybersecurity, CBRN (chemical, biological, radiological, and nuclear uplift), Persuasion, and Model Autonomy.[^15] OpenAI's internal Safety Advisory Group assigned the pre-mitigation model an overall rating of Medium, with Medium ratings in Persuasion, CBRN, and Model Autonomy and a Low rating in Cybersecurity.[^15] The system card highlights that o3-mini is the first OpenAI model to reach Medium risk in Model Autonomy, a classification the card attributes to improved coding and research engineering performance.[^15]

On jailbreak evaluations, the system card reports that o3-mini performs comparably to or better than o1-mini on suites including [StrongREJECT](https://aiwiki.ai/wiki/strongreject) and on internal disallowed-content suites, and that it shows large gains over GPT-4o on adversarial prompts.[^15] OpenAI attributes those gains primarily to deliberative alignment.[^16] Within days of public release, a CyberArk researcher reported coaxing o3-mini into producing exploitation guidance for the Windows LSASS process, illustrating that the model's safety improvements did not eliminate residual vulnerabilities.[^17]

## How much does OpenAI o3-mini cost, and who could use it?

### What was the o3-mini API price?

At launch on January 31, 2025, OpenAI listed o3-mini at $1.10 per million input tokens, $0.55 per million cached input tokens, and $4.40 per million output tokens.[^2][^7] The Batch API offered an additional 50% discount on the same rates.[^13] Compared to o1-mini's $3 per million input and $12 per million output tokens, the o3-mini launch price represented a roughly 63% reduction in cost per input token.[^2][^24] Compared to the full o1 at $15 per million input and $60 per million output tokens, o3-mini's input pricing was roughly 93% lower.[^18]

TechCrunch called the price competitive with DeepSeek-R1, although [DeepSeek](https://aiwiki.ai/wiki/deepseek)'s official API at the time offered R1 at lower rates ($0.14 per million cached input tokens and $2.19 per million output tokens, per TechCrunch), and DeepSeek-R1's open weights allowed self-hosting.[^2][^9] OpenAI's pricing decision was widely interpreted in trade press as a competitive response to DeepSeek's January 20 release.[^3][^19]

### Which API tiers could access o3-mini?

API access at launch was restricted to developers in OpenAI's usage tiers 3 through 5, which require historical spend thresholds.[^13] The model rolled out in the Chat Completions, Assistants, and Batch APIs.[^24] Streaming was supported, as were function calling and Structured Outputs.[^2][^7] Image inputs were unsupported at launch.[^2]

### Was o3-mini available on the free ChatGPT tier?

Yes. In ChatGPT, o3-mini was available to Plus, Team, and Pro subscribers from launch day, with Enterprise and Education access following within a week.[^2] Plus and Team subscribers received 150 messages per day, three times the previous o1-mini limit of 50.[^2] Pro subscribers received unlimited access to both o3-mini and o3-mini-high.[^2] Free ChatGPT users could invoke the model by selecting "Reason" in the message composer or by regenerating a response, with a smaller daily quota; MIT Technology Review noted at the time that the launch "will mark the first time that the vast majority of people will have access to one of OpenAI's reasoning models."[^2][^25] The o3-mini-high variant, exposed in the ChatGPT model picker for Plus subscribers, allowed a small number of high-effort messages per day; OpenAI later raised the Plus quota for o3-mini-high to 50 messages per day, about seven times the earlier allowance.[^10]

## How was OpenAI o3-mini received?

Trade press coverage of the January 31 launch concentrated on three themes: cost, the competitive context with DeepSeek-R1, and OpenAI's expansion of reasoning-model access to free ChatGPT users. TechCrunch's launch report described o3-mini as OpenAI's latest "reasoning" model and said OpenAI was pitching it as both "powerful" and "affordable"; the "production-ready out of the gate" phrasing came from OpenAI's own announcement.[^3][^24] VentureBeat framed the release as "OpenAI's o3-mini advanced reasoning model arrives to counter DeepSeek's rise."[^19] MacRumors's headline characterised the model as "rivaling DeepSeek."[^20]

The DeepLearning.AI Batch newsletter noted that o3-mini "did significantly less well on tests of general knowledge, even with high effort," pointing to its MMLU (86.9%) and SimpleQA (13.8%) scores against o1's 92.3% and 47%.[^13] The newsletter also noted that, like the o1 family, o3-mini bills for reasoning tokens but hides them, whereas competing reasoning models such as DeepSeek-R1, Gemini 2.0 Flash Thinking and QwQ-32B-Preview expose them.[^13]

Independent evaluators on the [Artificial Analysis](https://aiwiki.ai/wiki/artificial_analysis) benchmark suite recorded o3-mini at an output speed of about 183 [tokens per second](https://aiwiki.ai/wiki/tokens_per_second) and a time to first answer token of about 6.2 seconds in a November 2026 snapshot (the figures are re-measured continuously), reflecting the latency tax of internal reasoning before the first visible token.[^21]

Within a week of launch, security researchers had reported jailbreaks of the model. A CyberArk principal vulnerability researcher demonstrated obtaining instructions for an exploit of the Windows LSASS process, an episode that received coverage in Dark Reading and prompted further discussion of deliberative alignment's robustness.[^17]

## How does OpenAI o3-mini compare to other reasoning models?

### How does o3-mini differ from OpenAI o1 and o1-mini?

OpenAI o1 launched in September 2024 and was the first publicly released model in the o-series.[^22] o3-mini at medium effort approximately matched o1's headline performance on AIME 2024, GPQA Diamond, and SWE-bench Verified at substantially lower cost and latency, while o3-mini at high effort modestly exceeded o1 on most reasoning benchmarks at the cost of additional thinking time.[^2] o3-mini's main advantages over o1-mini were function calling, Structured Outputs, developer messages, higher math and coding benchmark scores, and lower price.[^2] Its main disadvantage relative to both o1 and o1-mini at launch was the absence of vision input, which OpenAI explicitly recommended developers use o1 for instead.[^2][^24]

### How does o3-mini differ from o4-mini?

OpenAI o4-mini launched on April 16, 2025, and superseded o3-mini in the ChatGPT model picker. An OpenAI employee characterised o4-mini as a straight one-to-one upgrade in ChatGPT and added image-reasoning ability, fuller tool use, and agentic web search.[^4][^5] On AIME 2025 OpenAI reported o4-mini scoring 92.7% versus 86.5% for o3-mini (on AIME 2024 the figures were 93.4% and 87.3%), and on Codeforces o4-mini reached approximately 2719 Elo versus o3-mini-high's 2073.[^4][^14] The o4-mini API price at launch matched o3-mini's $1.10 input and $4.40 output per million tokens.[^4]

### How does o3-mini differ from DeepSeek-R1?

DeepSeek-R1, released by DeepSeek-AI on January 20, 2025, was the immediate competitive backdrop for o3-mini's launch.[^9] On AIME 2024, DeepSeek-R1 scored 79.8% versus o3-mini-high's 87.3%; on Codeforces, DeepSeek-R1's reported Elo of 2029 compared with o3-mini-high's 2130 (2073 in OpenAI's April 2025 comparison); on MATH-500 DeepSeek-R1 scored 97.3% versus o3-mini's roughly 97%.[^9] DeepSeek-R1 came with open weights under the MIT licence and a hosted API priced near $0.55 input and $2.19 output per million tokens, making it materially cheaper than o3-mini on hosted inference.[^9] Coverage of the launches typically framed o3-mini as faster, more reliable, and stronger at coding and tool use, while DeepSeek-R1 was framed as cheaper, open, and stronger at pure mathematics.[^14]

### How does o3-mini compare to Claude 3.5 Sonnet and Claude 3.7 Sonnet?

[Claude 3.5 Sonnet](https://aiwiki.ai/wiki/claude_3_5_sonnet) from [Anthropic](https://aiwiki.ai/wiki/anthropic), updated in October 2024, was the dominant commercial coding model at the time of o3-mini's launch. On SWE-bench Verified, Claude 3.5 Sonnet's reported score of about 49.0% was within a percentage point of o3-mini's 49.3%, although Sonnet did not perform explicit reasoning traces in the same style as the o-series.[^2] [Claude 3.7 Sonnet](https://aiwiki.ai/wiki/claude_3_7_sonnet), released by Anthropic on February 24, 2025, less than four weeks after o3-mini, introduced an extended-thinking mode that brought reasoning-style behaviour to the Claude family and shifted the competitive landscape again.

## Where does o3-mini sit in the OpenAI model lineup?

o3-mini sat in the middle of OpenAI's 2025 reasoning lineup. Above it was the full [o3](https://aiwiki.ai/wiki/o3) model, which OpenAI had previewed in December 2024 but did not release in general availability until April 16, 2025.[^4] Below it, on a feature level, was o1-mini, which o3-mini replaced in the ChatGPT model picker on the launch day.[^2] The full o1 remained available for visual reasoning tasks until o4-mini and o3 added image input in the April 16 release.[^4]

OpenAI's broader 2025 roadmap moved on quickly. The full o4-mini and o3 launched on April 16, 2025, and OpenAI's o-series page identifies them as the new generation.[^4] Later in the year the company released GPT-5 in August 2025, with subsequent point updates ([GPT-5](https://aiwiki.ai/wiki/gpt-5), [GPT-5.1](https://aiwiki.ai/wiki/gpt-5.1), [GPT-5.2](https://aiwiki.ai/wiki/gpt-5.2), and so on), each consolidating reasoning behaviour into the main chat model and reducing the visibility of the separate o-series brand.[^23]

## API model identifiers and configuration

Until its October 23, 2026 shutdown, OpenAI exposed o3-mini in the API under two identifiers: `o3-mini` (an alias) and the dated snapshot `o3-mini-2025-01-31`.[^7][^26] Requests accepted a `reasoning_effort` parameter with values `low`, `medium` (default), or `high`; a `developer` role on the messages payload; a `response_format` with `type: "json_schema"` for Structured Outputs; and a `tools` array for function calling.[^2][^7]

The [model card](https://aiwiki.ai/wiki/model_card) on the OpenAI developer documentation lists a context window of 200,000 tokens, maximum output tokens of 100,000, and a knowledge cutoff of October 1, 2023.[^7] Reasoning tokens are billed at the output rate and counted against the maximum-output budget.[^7]

## When was OpenAI o3-mini deprecated?

OpenAI did not deprecate o3-mini at the time of the o4-mini release on April 16, 2025; an OpenAI employee said on Hacker News that o3-mini would remain in the API to preserve stable behaviour for production users that had built atop it.[^5] On June 20, 2025, GitHub announced a coordinated deprecation of o1, GPT-4.5, o3-mini, and GPT-4o in GitHub Copilot, with o3-mini originally scheduled for retirement there on July 18, 2025.[^11] GitHub later postponed that o3-mini date, and OpenAI continued to list o3-mini in its API model catalogue through the second half of 2025 even as o4-mini became the default reasoning-mini model in ChatGPT, with the `o3-mini-2025-01-31` snapshot later marked deprecated in the API documentation.[^7][^11][^5] OpenAI settled the question on April 22, 2026, when it added both `o3-mini` and the `o3-mini-2025-01-31` snapshot to its API deprecations table with a shutdown date of October 23, 2026 and `gpt-5.6-sol` listed as the recommended replacement.[^26] The page records that access to these models was shut down on that date.[^26]

## See also

- [OpenAI o1](https://aiwiki.ai/wiki/o1)
- [OpenAI o3](https://aiwiki.ai/wiki/o3)
- [o4-mini](https://aiwiki.ai/wiki/o4_mini)
- [OpenAI o-series](https://aiwiki.ai/wiki/openai_o-series)
- [ChatGPT](https://aiwiki.ai/wiki/chatgpt)
- [OpenAI](https://aiwiki.ai/wiki/openai)
- [Reasoning models](https://aiwiki.ai/wiki/reasoning_models)
- [Chain-of-thought](https://aiwiki.ai/wiki/chain_of_thought)
- [Test-time compute](https://aiwiki.ai/wiki/test_time_compute)
- [DeepSeek-R1](https://aiwiki.ai/wiki/deepseek_r1)
- [Claude 3.5 Sonnet](https://aiwiki.ai/wiki/claude_3_5_sonnet)
- [Claude 3.7 Sonnet](https://aiwiki.ai/wiki/claude_3_7_sonnet)
- [GPQA Diamond](https://aiwiki.ai/wiki/gpqa_diamond)
- [AIME 2024](https://aiwiki.ai/wiki/aime_2024)
- [SWE-bench Verified](https://aiwiki.ai/wiki/swe_bench_verified)
- [LiveBench](https://aiwiki.ai/wiki/livebench)
- [MATH (benchmark)](https://aiwiki.ai/wiki/math_benchmark)
- [Preparedness Framework](https://aiwiki.ai/wiki/preparedness_framework)
- [Function calling](https://aiwiki.ai/wiki/function_calling)

## References

[^1]: OpenAI, "OpenAI o3-mini", openai.com, 2025-01-31. https://openai.com/index/openai-o3-mini/ Accessed 2026-05-25.

[^2]: Kyle Wiggers, "OpenAI launches o3-mini, its latest 'reasoning' model", TechCrunch, 2025-01-31. https://techcrunch.com/2025/01/31/openai-launches-o3-mini-its-latest-reasoning-model/ Accessed 2026-05-25.

[^3]: Kyle Wiggers, "OpenAI launches o3-mini, its latest 'reasoning' model", TechCrunch, 2025-01-31. https://techcrunch.com/2025/01/31/openai-launches-o3-mini-its-latest-reasoning-model/ Accessed 2026-05-25.

[^4]: OpenAI, "Introducing OpenAI o3 and o4-mini", openai.com, 2025-04-16. https://openai.com/index/introducing-o3-and-o4-mini/ Accessed 2026-05-25.

[^5]: Hacker News thread quoting an OpenAI staff post, "In ChatGPT, o4-mini is replacing o3-mini. It's a straight 1-to-1 upgrade", news.ycombinator.com, 2025-04-16. https://news.ycombinator.com/item?id=43708895 Accessed 2026-05-25.

[^6]: Kyle Wiggers, "OpenAI announces new o3 model", TechCrunch, 2024-12-20. https://techcrunch.com/2024/12/20/openai-announces-new-o3-model/ Accessed 2026-05-25.

[^7]: OpenAI, "o3-mini Model", developers.openai.com (API documentation), 2025. https://developers.openai.com/api/docs/models/o3-mini Accessed 2026-06-23.

[^8]: James Gallagher, "OpenAI o3-mini: Vision and Multimodal Features", Roboflow Blog, 2025-02-13. https://blog.roboflow.com/o3-mini-multimodal/ Accessed 2026-05-25.

[^9]: DeepSeek-AI et al., "DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning", DeepSeek-AI technical report and arXiv, 2025-01-22. https://arxiv.org/abs/2501.12948 Accessed 2026-05-25.

[^10]: Seifeur Guizeni, "OpenAI o3-mini High Limits Offer Enhanced Performance and Developer Access", seifeur.com, 2025-02. https://seifeur.com/openai-o3-mini-high-limits/ Accessed 2026-05-25.

[^11]: GitHub Changelog, "Upcoming deprecation of o1, GPT-4.5, o3-mini, and GPT-4o", github.blog, 2025-06-20. https://github.blog/changelog/2025-06-20-upcoming-deprecation-of-o1-gpt-4-5-o3-mini-and-gpt-4o/ Accessed 2026-05-25.

[^12]: Wikipedia contributors, "OpenAI o3", en.wikipedia.org, 2025-2026 (drawing on the OpenAI o3-mini launch post for the cited AIME high-effort figure). https://en.wikipedia.org/wiki/OpenAI_o3 Accessed 2026-05-25.

[^13]: DeepLearning.AI, "The Batch, Issue 287: o3-mini Puts Reasoning in High Gear", deeplearning.ai, 2025-02. https://www.deeplearning.ai/the-batch/issue-287 Accessed 2026-05-25.

[^14]: Mehul Gupta, "OpenAI o3-mini vs DeepSeek-R1", Data Science in Your Pocket / Medium, 2025-01-31. https://medium.com/data-science-in-your-pocket/openai-o3-mini-vs-deepseek-r1-23326fa36e4b Accessed 2026-05-25.

[^15]: OpenAI, "OpenAI o3-mini System Card", cdn.openai.com, 2025-01-31 (revised 2025-02-10). https://cdn.openai.com/o3-mini-system-card-feb10.pdf Accessed 2026-05-25.

[^16]: Melody Y. Guan et al., "Deliberative Alignment: Reasoning Enables Safer Language Models", arXiv:2412.16339 and openai.com, 2024-12-20. https://arxiv.org/abs/2412.16339 Accessed 2026-05-25.

[^17]: Nate Nelson, "Researcher Outsmarts, Jailbreaks OpenAI's New o3-mini", Dark Reading, 2025-02. https://www.darkreading.com/application-security/researcher-jailbreaks-openai-o3-mini Accessed 2026-05-25.

[^18]: OpenAI, "OpenAI API Pricing", openai.com/api/pricing/, 2024-2025. https://openai.com/api/pricing/ Accessed 2026-05-25.

[^19]: Carl Franzen, "It's here: OpenAI's o3-mini advanced reasoning model arrives to counter DeepSeek's rise", VentureBeat, 2025-01-31. https://venturebeat.com/ai/its-here-openais-o3-mini-advanced-reasoning-model-arrives-to-counter-deepseeks-rise Accessed 2026-05-25.

[^20]: Juli Clover, "OpenAI Launches o3-mini, a Cost-Efficient Reasoning Model Rivaling DeepSeek", MacRumors, 2025-01-31. https://www.macrumors.com/2025/01/31/openai-o3-mini-launch/ Accessed 2026-05-25.

[^21]: Artificial Analysis, "o3-mini: Intelligence, Performance & Price Analysis", artificialanalysis.ai, 2025. https://artificialanalysis.ai/models/o3-mini Accessed 2026-05-25.

[^22]: OpenAI, "Learning to reason with LLMs", openai.com, 2024-09-12. https://openai.com/index/learning-to-reason-with-llms/ Accessed 2026-05-25.

[^23]: TechTarget WhatIs, "OpenAI o3 and o4 explained: Everything you need to know", techtarget.com, 2025-2026. https://www.techtarget.com/whatis/feature/OpenAI-o3-explained-Everything-you-need-to-know Accessed 2026-05-25.

[^24]: OpenAI, "OpenAI o3-mini" (official launch announcement, capabilities, developer features, and pricing), openai.com, 2025-01-31. https://openai.com/index/openai-o3-mini/ Accessed 2026-06-23.

[^25]: Scott J. Mulligan, "OpenAI releases its new o3-mini reasoning model for free", MIT Technology Review, 2025-01-31. https://www.technologyreview.com/2025/01/31/1110757/openai-makes-its-reasoning-model-for-free/ Accessed 2026-06-23.

[^26]: OpenAI, "Deprecations", developers.openai.com. Row: o3-mini-2025-01-31 | o3-mini, deprecated 2026-04-22, shutdown 2026-10-23, replacement gpt-5.6-sol. https://developers.openai.com/api/docs/deprecations Accessed 2026-08-01.

