benchmarks
GPT-5.1 is an OpenAI model supporting text and image inputs, text output, configurable reasoning, and tool-assisted workflows. OpenAI subsequently released GPT-5.2; the specifications and results here describe GPT-5.1 rather than its successors or Codex variants.
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 | 87.6%13 Nov 2025 | View |
| SWE-bench VerifiedResolving real GitHub issues | 68%18 Feb 2026 | View |
| OTIS Mock AIME 2024-25Competition mathematics | 88.6%13 Nov 2025 | View |
| Terminal-BenchLong-horizon command-line agent tasks | 47.6%16 Nov 2025 | View |
| FrontierMath | 31%13 Nov 2025 | View |
| SimpleQA VerifiedShort factual questions (hallucination resistance) | 48%27 Aug 2026 | View |
| LMArena TextHuman preference Elo from blind chat votes (not a percentage) | 1443 Elo22 Oct 2026 | View |
| LMArena WebDevHuman preference Elo for building web apps (not a percentage) | 1396 Elo23 Oct 2026 | View |
Weighted mean of 6 verified benchmark results: GPQA Diamond 87.6%, SWE-bench Verified 68%, OTIS Mock AIME 2024-25 88.6%, Terminal-Bench 47.6%, FrontierMath 31%, SimpleQA Verified 48%.
Capabilities
- Processes text and image inputs and generates text responses.
- Supports configurable reasoning effort, including a no-reasoning setting.
- Supports function calling and structured outputs for application integration.
- Handles long-context document analysis within a 400,000-token context window.
- Supports coding workflows, including repository changes, debugging, and tool-assisted development.
What improved
- Allocates reasoning more adaptively than GPT-5, reducing unnecessary processing on simpler tasks.
- Adds a no-reasoning setting for lower-latency interactions.
- Supports extended prompt caching with retention up to 24 hours.
- Improves vendor-reported SWE-bench Verified performance from GPT-5's 72.8% to 76.3% under OpenAI's evaluation setup.
- Improves coding steerability and communication during development workflows.
Known flaws & limitations
- The API model does not natively accept audio or video input.
- Factual responses and generated code can be incorrect and require verification.
- Function calls require application-side execution; model output is not a secure execution environment.
- Benchmark performance depends on reasoning settings, tools, and evaluation configuration.
- GPT-5.1-Codex-Max results and long-running coding claims do not describe the base GPT-5.1 model.
Risks, safety & security issues
- Untrusted documents, retrieved content, and tool outputs can introduce prompt-injection risks in agent workflows.
- Generated commands or code can cause unintended changes when applications grant excessive execution permissions.
- Sensitive information in context can be exposed through inadequately controlled responses or tool interactions.
- Safety refusals and jailbreak defenses are not substitutes for application-level authorization and security controls.
Good fit for
- Coding assistance and repository maintenance with testing and review.
- Long-document analysis and synthesis with source verification.
- Tool-assisted business workflows with explicit authorization boundaries.
- Interactive applications needing configurable reasoning and latency.
Poor fit for
- Native audio or video processing.
- Unsupervised execution of consequential commands with broad permissions.
- High-stakes decisions requiring guaranteed factual accuracy.
- Workloads selected on the assumption that successor or Codex benchmark results apply to GPT-5.1.
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.