benchmarks
OpenAI o4-mini is a small reasoning model supporting text and image inputs, text outputs, and tool-assisted workflows. GPT-5 mini is a newer successor; o4-mini's documented standard API pricing is $1.10 per million input tokens and $4.40 per million output tokens.
Benchmark results
Every figure links to where it was published. Compare like with like — harnesses and prompt setups differ between vendors.
| Benchmark | Score | Source |
|---|---|---|
| GPQA DiamondPhD-level science questions | 79.6%16 Apr 2025 | View |
| OTIS Mock AIME 2024-25Competition mathematics | 81.7%16 Apr 2025 | View |
| FrontierMath | 24.8%13 Nov 2025 | View |
| SimpleQA VerifiedShort factual questions (hallucination resistance) | 19.6%27 Aug 2026 | View |
| MATH Level 5 | 97.8%16 Apr 2025 | View |
| LMArena TextHuman preference Elo from blind chat votes (not a percentage) | 1353 Elo30 Sep 2026 | View |
| Aider PolyglotCode editing across six programming languages | 72%16 Apr 2025 | View |
| LiveCodeBenchFresh competitive-programming problems | 87.3%1 May 2025 | View |
Weighted mean of 7 verified benchmark results: GPQA Diamond 79.6%, OTIS Mock AIME 2024-25 81.7%, FrontierMath 24.8%, SimpleQA Verified 19.6%, MATH Level 5 97.8%, Aider Polyglot 72%, LiveCodeBench 87.3%.
Capabilities
- Multi-step reasoning for mathematics, science and programming.
- Image-input reasoning for diagrams, charts and other visual material.
- Function calling and structured outputs for application integration.
- Configurable reasoning effort to balance computational work and responsiveness.
- Long-context processing with a 200,000-token context window.
- Tool-assisted workflows combining search, Python and visual analysis where the hosting product provides those tools.
What improved
- Adds image-input reasoning compared with the text-only o3-mini.
- Achieves higher published AIME 2024 and AIME 2025 results than o3-mini.
- Achieves higher published SWE-bench Verified performance than o3-mini.
- Introduced broader tool orchestration in ChatGPT, combining web search, Python and visual analysis.
Known flaws & limitations
- Can produce confident factual errors; the system card reports higher hallucination rates than o3-mini on evaluated factuality tasks.
- Does not support native audio or video input.
- Produces text rather than native image or audio output.
- Reasoning tokens consume the output budget, incur output-token charges and can increase latency.
- Published benchmark results depend on reasoning effort, tool access and evaluation configuration.
Risks, safety & security issues
- Malicious instructions embedded in untrusted content can redirect tool-connected workflows.
- Residual jailbreak and harmful-assistance risks remain despite safety training.
- Confident factual errors can mislead users making consequential decisions.
- Tool access can turn mistaken reasoning into external actions unless permissions and approval gates constrain execution.
Good fit for
- Cost-conscious mathematics and science problem solving with independently checked answers.
- Coding assistance, debugging and repository tasks backed by automated tests.
- Reasoning over charts, diagrams and mixed text-image documents.
- Tool-connected applications with restricted permissions and human approval for consequential actions.
Poor fit for
- Unsupervised medical, legal or financial decisions requiring reliably accurate facts.
- Native audio, video or image-generation workloads.
- Strict low-latency workloads that cannot accommodate variable reasoning time.
- Autonomous workflows with unrestricted access to sensitive data or irreversible actions.
References
Profile last reviewed 10/27/2026 and re-checked automatically every few hours against the vendor's published documentation, papers and public leaderboards. Scores come from the cited benchmark results above — confirm limits and pricing against the provider's own documentation before building on them.