benchmarks
GPT-4.1 mini is OpenAI’s compact GPT-4.1 model for text and image understanding, instruction following, and tool-assisted applications. Its documented specifications include a roughly one-million-token context window and standard API pricing of $0.40 per million input tokens and $1.60 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 | 65.8%14 Apr 2025 | View |
| OTIS Mock AIME 2024-25Competition mathematics | 44.7%14 Apr 2025 | View |
| FrontierMath | 4.5%14 Apr 2025 | View |
| SimpleQA VerifiedShort factual questions (hallucination resistance) | 12.7%31 Aug 2026 | View |
| MATH Level 5 | 87.3%14 Apr 2025 | View |
| Aider PolyglotCode editing across six programming languages | 32.4%14 Apr 2025 | View |
Weighted mean of 6 verified benchmark results: GPQA Diamond 65.8%, OTIS Mock AIME 2024-25 44.7%, FrontierMath 4.5%, SimpleQA Verified 12.7%, MATH Level 5 87.3%, Aider Polyglot 32.4%.
Capabilities
- Accepts text and image inputs and generates text outputs.
- Not yet confirmed by public reporting.
- Supports function calling for applications that connect the model to external tools.
- Supports structured outputs for schema-constrained responses.
- Generates and edits code, with published software-engineering evaluations.
What improved
- Expands the context window from GPT-4o mini’s 128,000 tokens to 1,047,576 tokens.
- Increases maximum output length from GPT-4o mini’s 16,384 tokens to 32,768 tokens.
- Improves instruction following relative to GPT-4o mini in OpenAI’s launch evaluations.
- Improves scientific reasoning and multimodal understanding relative to GPT-4o mini in OpenAI’s launch evaluations.
Known flaws & limitations
- Its June 2024 knowledge cutoff means recent facts require external sources.
- Output is limited to 32,768 tokens, substantially less than the input context capacity.
- It does not provide native audio input or output.
- Published software-engineering performance trails the full GPT-4.1 model.
- A large context window does not guarantee reliable retrieval or reasoning across every supplied detail.
Risks, safety & security issues
- Incorrect factual answers can mislead users, particularly when used without independent verification in consequential decisions.
- Generated code and patches can contain defects; benchmark performance does not establish production safety.
- Outdated knowledge can produce misleading guidance about changing events, policies, or requirements.
Good fit for
- High-volume document summarization, classification, and information extraction.
- Tool-connected assistants that require structured responses.
- Image-assisted document understanding and question answering.
- Routine code generation and editing with tests and human review.
Poor fit for
- Native real-time audio conversations.
- Unsupervised high-stakes decisions requiring consistently accurate factual judgments.
- Difficult autonomous software-engineering work without testing and review.
- Current-events answers without access to reliable external sources.
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.