GPT-5.1

// Overview

GPT-5.1 is OpenAI's flagship large language model in the GPT-5 (Flagship / Thinking) line, released on November 12, 2025 in ChatGPT and rolled out to the API for developers on November 13. It ships as two models: GPT-5.1 Instant — a warmer, more conversational default tuned for everyday chat and quick coding — and GPT-5.1 Thinking, a reasoning model aimed at harder, multi-step problems. It positions itself as a smarter, friendlier successor to GPT-5 rather than a clean-sheet model.

Its headline change is adaptive reasoning: GPT-5.1 decides how much time to spend thinking based on how hard the task looks, spending fewer tokens on straightforward prompts and more on complex ones. OpenAI reports it runs roughly 2-3x faster than GPT-5 on simple tasks while using far fewer tokens, and it adds a new 'none' reasoning-effort option that makes the model respond like a non-reasoning model for latency-sensitive use. ChatGPT also gained eight selectable personality presets and a warmer default tone.

For developers, GPT-5.1 carries a 400,000-token context window, up to 128,000 output tokens, a September 30, 2024 knowledge cutoff, and text-plus-image (vision) input. The API release adds two new built-in tools — apply_patch for structured, reliable code edits and a shell tool for proposing shell commands a developer runs — plus extended prompt caching that retains prompts for up to 24 hours. It is priced at $1.25 per million input tokens and $10 per million output tokens, with cached input at $0.125 per million.

// Benchmarks

1. SWE-bench Verified — real-world software engineering76.3%
2. GPQA Diamond — PhD-level science reasoning88.1%
3. AIME 2025 — competition math94%
4. SWE-bench Pro — harder multi-file engineering50.8%
5. Artificial Analysis Intelligence Index (GPT-5.1 high)39

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

Standard API pricing for the gpt-5.1 model. Cached input is a 90% discount on input; extended prompt caching retains prompts for up to 24 hours. Reasoning tokens in Thinking mode are billed as output tokens.

Pricing source ↗

// Strengths

• Adaptive reasoning that spends fewer thinking tokens on easy prompts and more on hard ones — roughly 2-3x faster than GPT-5 on simple tasks at lower token cost
• Two complementary variants from one line: GPT-5.1 Instant (warm, fast, default) and GPT-5.1 Thinking (deeper reasoning for complex problems)
• A 'none' reasoning-effort setting that turns off thinking entirely for latency-sensitive, interactive use
• Strong coding and agentic results — 76.3% on SWE-bench Verified — plus new apply_patch and shell tools that tighten the edit-and-run loop
• Frontier science reasoning: 88.1% on GPQA Diamond and 94.0% on AIME 2025
• Large 400K-token context window with up to 24-hour extended prompt caching for cheaper, faster follow-ups
• Vision input (text + images) for screenshots, charts, diagrams, and document pages
• Warmer, more steerable ChatGPT personality with eight selectable presets

// Best for

• Everyday conversational assistance and writing where a warmer, faster default tone matters
• Agentic coding workflows that edit files (apply_patch) and propose shell commands for a developer to run
• Complex reasoning, math, and science tasks routed to GPT-5.1 Thinking
• Latency-sensitive interactive apps using Instant with reasoning_effort set to 'none'
• Long-document and large-codebase analysis within the 400K-token context window
• Vision-assisted steps that read screenshots, charts, scanned pages, and diagrams
• Cost-sensitive high-volume API workloads that benefit from adaptive token spend and 24-hour prompt caching

// How to access

// GPT (Flagship / Thinking) — every version

The full lineage of the GPT (Flagship / Thinking) line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.