benchmarks
GPT-5 mini is a smaller, lower-cost GPT-5 reasoning model for well-defined tasks, supporting text and image inputs with text output. It offers a 400,000-token context window and tool integration; newer GPT-5 family models have shipped, but this profile covers GPT-5 mini specifically.
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 | 75%30 Oct 2025 | View |
| SWE-bench VerifiedResolving real GitHub issues | 64.7%1 Feb 2026 | View |
| OTIS Mock AIME 2024-25Competition mathematics | 86.7%30 Oct 2025 | View |
| Terminal-BenchLong-horizon command-line agent tasks | 34.8%7 Aug 2025 | View |
| FrontierMath | 27.2%13 Nov 2025 | View |
| SimpleQA VerifiedShort factual questions (hallucination resistance) | 21.6%10 Aug 2026 | View |
| MATH Level 5 | 97.8%30 Oct 2025 | View |
| LMArena TextHuman preference Elo from blind chat votes (not a percentage) | 1373 Elo22 Oct 2026 | View |
Weighted mean of 7 verified benchmark results: GPQA Diamond 75%, SWE-bench Verified 64.7%, OTIS Mock AIME 2024-25 86.7%, Terminal-Bench 34.8%, FrontierMath 27.2%, SimpleQA Verified 21.6%, MATH Level 5 97.8%.
Capabilities
- Accepts text and image inputs and generates text responses.
- Supports a 400,000-token context window and up to 128,000 output tokens.
- Provides configurable reasoning effort and response verbosity.
- Supports function calling and structured outputs for application integration.
- Handles coding, mathematical reasoning, and instruction-following tasks at standard API prices of $0.25 per million input tokens and $2 per million output tokens.
What improved
- Compared with GPT-4o mini, expands the context window from 128,000 to 400,000 tokens.
- Compared with GPT-4o mini, increases the maximum output allowance from 16,384 to 128,000 tokens.
- Adds configurable reasoning effort and response-verbosity controls relative to GPT-4o mini.
Known flaws & limitations
- Generates text only; it does not generate images, audio, or video.
- Does not directly support audio or video input.
- Reasoning tokens count toward output capacity and are billed even when they do not appear in the visible answer.
- Can produce factual errors and faulty reasoning; benchmark performance does not guarantee individual answers.
Risks, safety & security issues
- Prompt injection in untrusted documents or tool results can attempt to redirect tool-enabled applications.
- Incorrect but persuasive answers can cause harm when used without independent review in consequential decisions.
- Excessive tool permissions can turn model mistakes or malicious instructions into sensitive-data exposure or unintended actions.
Good fit for
- High-volume classification, extraction, and structured-response workflows with clear instructions.
- Coding assistance and repository tasks with automated tests and human review.
- Long-document analysis and question answering over supplied text.
- Text-and-image analysis that requires a text response.
Poor fit for
- Native audio or video processing, or image, audio, and video generation.
- Unsupervised medical, legal, financial, or other high-stakes decision-making.
- Autonomous workflows with unrestricted access to sensitive data or irreversible actions.
References
Profile last reviewed 10/25/2026 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.