HomeAI models
OpenAI: o4 Mini
OpenAI
•
o4-mini (High)
•
Released 
April 2025

OpenAI
o4-mini (High)

Small OpenAI reasoning model with 100K-token outputs, image input and fast generation, now approaching its scheduled October 2026 shutdown.

Modality:
Text
Image
Video
PDF
model ID
openai/o4-mini
Start for free
arrow black icon
View API Docs
arrow white icon

Output Speed *

N/A
tok/s

Intelligence Index *

16.7
/ 100

Context Window *

200000
tokens

Input price

6.6
Anytoken

Output price

26.4
Anytoken
o4-mini (High): Fast Reasoning With Image Input and 100K-Token Outputs o4-mini (High) is OpenAI's small o-series reasoning model run at its high reasoning-effort setting. Released on April 16, 2025 alongside o3, it sits below o3 in the lineup and targets math, coding and visual tasks. It accepts text and image input, returns text, and offers a 200,000-token context window with up to 100,000 output tokens. It suits reasoning-heavy, high-volume jobs where output speed matters. OpenAI has scheduled the o4-mini API shutdown for October 23, 2026, so new builds should plan migration. Test o4-mini (High) on AnyAPI.ai

Performance

Math and Coding Reasoning at Small-Model Speed

o4-mini was built for math, coding and visual reasoning at lower cost than o3. OpenAI reported 92.7% on AIME 2025 without tools and 68.1% on SWE-bench Verified, close to o3's 69.1%. That puts a smaller, faster model within about one point of its larger sibling on a coding-agent benchmark. In production, it fits high-volume reasoning calls where o3-level answers are not required. The high setting adds hidden reasoning tokens and latency, so budget output tokens accordingly.

Benchmarks

Independent Data: High Output Speed, Long Wait for the First Token

Artificial Analysis measured o4-mini (High) at about 155 tokens per second on OpenAI's API, roughly double the 77.7 median for comparable reasoning models in a similar price tier. The trade-off is latency. Its time to first token was 30.79 seconds in a head-to-head with GPT-4.1, which returned a first token in 1.05 seconds. In that comparison, o4-mini (High) led on Humanity's Last Exam (17% vs 4%) but trailed on AA-LCR long-context reasoning (61% vs 68%). Figures shift between snapshots.

Output Speed

*
N/A
tok/s

Intelligence Index

*
16.7
/ 100

MMLU *

Broad world knowledge and problem-solving
83
%

GPQA *

PhD-level scientific reasoning across physics, biology, chemistry.
78
%

HLE *

Adherence to multi-step structured instructions.
17
%

LiveCodeBench *

Tool-calling reliability in long agentic loops.
86
%

Technical Specifications

What the model supports

o4-mini accepts text and image input and returns text only. Audio and video are not supported. The 200,000-token context window is separate from the 100,000-token output ceiling, and reasoning tokens at high effort count toward that output budget. Function calling, structured outputs and streaming are supported; predicted outputs are not. For most calls, the practical constraint is reasoning overhead and latency rather than the output cap.
Verified Specifications — 
o4-mini (High)
*
Input modalities
Text
Image
Video
PDF
output modalities
Text
Context window
200000
 tokens
Maximum output tokens
100000
Reasoning
Yes
Knowledge cutoff
April 2025
Pricing (standard)
6.6
 AnyTokens in
 / 
26.4
 AnyTokens out

Comparison

o4-mini (High) or GPT-5 mini: Is Migrating Worth It?

OpenAI's model page names GPT-5 mini as the successor to o4-mini, and both are small, cost-efficient reasoning models. In Artificial Analysis testing at high reasoning settings, the two post the same Intelligence Index score of 17, with the o4-mini figure marked as estimated. o4-mini (High) generated about 142 tokens per second against roughly 98 for GPT-5 mini (high). With the o4-mini API shutdown scheduled for October 23, 2026, the decision is mostly migration timing versus raw output speed.

Dimension
o4-mini (High)
GPT-5 mini (High)
Context window *
200000
tokens
400000
tokens
Output speed *
N/A
tok/s
0.00
tok/s
Intelligence Index *
16.7
16.8
Input pricing
6.6
AnyToken
1.5
AnyToken
Output pricing
26.4
AnyToken
12
AnyToken
Knowledge cutoff *
April 2025
August 2025

Choose o4-mini (High) when you maintain an existing integration, value the faster output speed Artificial Analysis measured, and can finish before the shutdown date. Choose GPT-5 mini when starting a new build, when you want the model OpenAI documents as o4-mini's successor, or when you need a path not tied to the October 23, 2026 retirement. Re-test prompts, parameters and token budgets either way, because a replacement is not a guaranteed drop-in.

Limitations & Trade-offs

Where o4-mini (High) falls short

1
Scheduled API shutdown. OpenAI's deprecation schedule lists o4-mini and the o4-mini-2025-04-16 snapshot for shutdown on October 23, 2026, with gpt-5.6-terra named as the replacement. After that date, requests using the model ID are expected to fail. The model is a poor choice for new production systems, and existing workloads need migration testing now. Confirm current availability for your account in AnyAPI.ai before relying on it.
2
Long time to first token. Artificial Analysis measured 30.79 seconds to first token in one comparison, with other snapshots between roughly 15 and 23 seconds, because the model reasons before answering and high effort extends that. GPT-4.1 returned a first token in about one second in the same test. Chat interfaces, autocomplete and voice agents suffer. Use o4-mini (High) for background or latency-tolerant reasoning, and a non-reasoning model for live interaction.
3
Behind newer models on general intelligence. In Artificial Analysis head-to-heads, GPT-5 (high) scored 23 and DeepSeek V4 Pro scored 36, against an estimated 17 for o4-mini (High). The model is also below the median of comparable models in one snapshot. For hard agentic, research or long-context tasks, a newer model is usually the safer pick, and o4-mini's AA-LCR score trailed GPT-4.1 in the same comparison.
4
Narrow modality and verbose reasoning. Input is limited to text and images, with no audio or video, and output is text only. Predicted outputs are not supported. Artificial Analysis also noted that o4-mini (High) generated well above the median number of output tokens for similar-tier reasoning models during its index run. Expect higher token consumption on simple prompts than non-reasoning models would use.

Best-Fit Workloads

Where this model earns its place

01

Math and STEM problem solving

‍
OpenAI reported 92.7% on AIME 2025 without tools, 99.5% with a Python interpreter, and 81.4% on GPQA Diamond. This suits quantitative tutoring, solution checking and scientific analysis pipelines. Tool-assisted scores are not comparable to no-tool scores, and your application must supply the interpreter. Expect slow first tokens, so run these jobs asynchronously rather than in a live chat loop.

02

Bug-fixing and code-review agents

‍
OpenAI reported 68.1% on SWE-bench Verified for o4-mini, close to o3, using a 477-task subset. Function calling and structured outputs support agent loops that return machine-readable patches or review verdicts. OpenAI's run used a 256k context that exceeds the 200,000-token API window and lifted o4-mini's solve rate by about 3%, so expect slightly lower results on very large repositories.

03

Chart, diagram and screenshot analysis

‍
The model accepts image input alongside text, and OpenAI positioned it for visual tasks. Reported results include 81.6% on MMMU. It fits workflows such as reading charts in reports, interpreting UI screenshots for QA, or extracting reasoning from diagrams. Output is text only, and it cannot process audio or video, so those inputs need separate preprocessing.

04

High-volume background reasoning

‍
Artificial Analysis measured about 155 tokens per second, well above the median for similar-tier reasoning models, and the 100,000-token output cap allows long reports or multi-step analyses. This suits queued jobs like batch document analysis or report drafting where throughput matters more than first-token delay. The scheduled shutdown limits it to short-horizon or migration-period use.

Pricing in anytokens via AnyAPI
Input
6.6
₳
Output
26.4
₳
Cache write
—
₳
Cache read
1.65
₳
Get API access
arrow black icon
Read Docs

on this page

PerformanceBenchmarksTechnical SpecificationsComparisonLimitations & Trade-offsBest-Fit WorkloadsIntegrationFAQ

Integration

Access o4-mini (High) via AnyAPI.ai

Access o4-mini (High) through AnyAPI.ai using a unified API built for multi-model AI applications. Integrate o4-mini (High) without maintaining a separate provider-specific connection, and keep the flexibility to test, switch, or combine models as your application requirements evolve.

01

One API integration

Access o4-mini (High) and other AI models through the same API workflow instead of maintaining separate integrations for every provider.

02

Easy model switching

Test o4-mini (High) against alternative models or switch models as your performance, capability, or cost requirements change without rebuilding your application around another provider API.

03

Flexible for production

Use o4-mini (High) from experimentation through production while keeping your AI stack flexible as workloads, traffic, and model requirements evolve.

04

Multi-model applications

Use o4-mini (High) for the workloads where it performs best and combine it with other models for tasks that require different capabilities, performance, or efficiency.

Start free
arrow black icon
Read Docs

Frequently Asked Questions

Answers to common questions about integrating and using this AI model via AnyAPI.ai

o4-mini (High) has a 200,000-token context window and a maximum output of 100,000 tokens. The two limits are separate, and reasoning tokens count toward the output budget, so high-effort requests can use much of that ceiling before the visible answer. Input supports text and images; output is text only.

They are the same model. o4-mini (High) is o4-mini run with reasoning effort set to high, which spends more reasoning tokens for more accurate answers at the cost of higher latency and token usage. OpenAI introduced o4-mini-high as a ChatGPT option at launch; in the API it is controlled by the reasoning effort setting.

Yes. OpenAI's deprecation schedule lists o4-mini and the o4-mini-2025-04-16 snapshot for API shutdown on October 23, 2026, with gpt-5.6-terra as the named replacement. OpenAI's model page also names GPT-5 mini as its successor. Re-test prompts and parameters, since replacements are not guaranteed drop-ins.

Artificial Analysis measured about 155 tokens per second on OpenAI's API, roughly twice the median for comparable reasoning models. Time to first token was far slower, between roughly 15 and 31 seconds across snapshots, because the model reasons first. It suits throughput-heavy background work better than interactive replies.

OpenAI reported 68.1% for o4-mini on SWE-bench Verified, versus 69.1% for o3 and 49.3% for o3-mini, on a 477-task subset. These are vendor-reported figures. Artificial Analysis rates the model below newer models on its general Intelligence Index, so newer options may do better on harder agentic coding.

* Benchmark data source: Artificial Analysis artificialanalysis.ai