GPT-5.1
GPT-5.1 by OpenAI adds adaptive reasoning effort, apply_patch and shell tools, 24-hour prompt caching, and a no-reasoning mode for low-latency agentic work.
Model Overview
Capabilities, design details, and architectural traits
GPT-5.1 - Adaptive Reasoning Depth and Token-Efficient Thinking
GPT-5.1 is the version in the GPT 5 series where OpenAI redesigned the thinking loop to spend fewer tokens on simple tasks and more tokens on hard ones. This is not a routing decision made externally - it is a documented change to how the model was trained. On easy tasks, GPT-5.1 produces a correct answer in significantly fewer tokens than GPT 5; on hard tasks, it reasons longer and more persistently than GPT 5 did.
Two New Developer-Native Tools
GPT-5.1 ships with two tools introduced for the first time in the GPT 5 series:
apply_patch- lets the model emit structured diffs to create, update, or delete files in a codebase without JSON escaping, enabling iterative multi-step code editingshell- lets the model propose shell commands that an integration executes locally, then feeds the output back; creates an explicit plan-execute loop for filesystem and utility tasks
What Separates GPT-5.1
| Trait | Detail |
|---|---|
| Adaptive reasoning for Instant variant | First time GPT 5 Instant (non-reasoning) decides when to think before responding, based on task complexity - documented as a training change, not a prompt instruction |
reasoning_effort='none' as the explicit default | Defaults to no reasoning; documented as ideal for latency-sensitive workloads; developers opt into low, medium, or high as needed |
| 24-hour prompt cache retention | Extended from minutes to 24 hours via prompt_cache_retention='24h' on the Responses or Chat Completions API; designed for long-running sessions like multi-turn coding and knowledge retrieval |
| Eight ChatGPT personality presets | Introduced alongside the model: Default, Friendly, Efficient, Professional, Candid, Quirky, Cynical, and Nerdy - all steerable per conversation from personalization settings |
| Dedicated Codex variants | Releases gpt-5.1-codex and gpt-5.1-codex-mini as separate API models optimized specifically for long-running agentic coding harnesses like Codex |
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | OpenAI | Anthropic | Anthropic |
| Release Date | November 13, 2025 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | - | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 400K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | $1.25 Best Input Pricing | $4 | $2 |
| Output Pricing | $10 Best Output Pricing | $20 | $10 Best Output Pricing |
| Modalities | |||
| Inputs | imagetextfile | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 24.7 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | 49.4 | - | - |
| Agentic Index | 21.6 | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from OpenAI
Other models by OpenAI
Top AI Models
Leading alternatives by intelligence score
