openai/gpt-5.1

The best model for coding and agentic tasks with configurable reasoning effort.

84 runs
Playground API Examples README

GPT‑5.1

Two upgraded models from OpenAI’s GPT‑5 family, now available on Replicate:

  • gpt-5.1-chat-latest: Fast, conversational, and responsive. Ideal for everyday use.
  • gpt-5.1: Slower, more thorough, and better at complex reasoning.

These are drop-in replacements for GPT‑5 with improvements to instruction following, tone, and performance.

🧠 Models

gpt-5.1-chat-latest

The default option. Prioritizes speed, clarity, and natural conversation.

It handles casual questions, creative writing, and multiturn chats well. Compared to gpt-5, it follows instructions more accurately, especially in constrained formats like:

“respond in exactly six words”
“answer like a pirate”

The tone is warmer by default — less robotic, more human. That doesn’t mean less accurate, just easier to talk to.

Also supports adaptive reasoning: the model decides whether to think before answering. For simple queries, it replies quickly. For harder ones, it takes more time to get things right.

gpt-5.1

Built for harder tasks.

Spends more time on complex problems, like:

  • Writing technical explanations
  • Solving math problems
  • Summarizing dense research

On simpler prompts, it’s faster than older versions of GPT‑5 Thinking. On harder ones, it’s slower — by design.

It’s also better at staying on-topic and breaking down difficult ideas without assuming too much. Responses use clearer language and avoid jargon when possible.

✅ Improvements over GPT‑5

  • Tone: More natural and expressive
  • Instruction following: Better at formatting and edge cases
  • Adaptive reasoning: Balances speed and depth
  • Clarity: Simpler language for complex topics

💬 Example: Tone + Reasoning

Prompt:

I spilled coffee on myself before a meeting. Do you think everyone thought I looked stupid?

gpt-5.1-chat-latest:

You’re rattled, so your brain is catastrophizing. Most people didn’t notice — they’re juggling their own chaos. A coffee spill reads as “busy human,” not “idiot.”

gpt-5.1:

You’re not an idiot. You’re a person who had a chaotic moment, wiped it up, and still showed up. That’s resilience. Not failure.

🚀 Availability

  • gpt-5.1-chat-latest: available now
  • gpt-5.1: available now

Both models support streamed output and system/instruction messages.

🏷️ Naming

These models are part of the GPT‑5 generation. The .1 indicates a mid-cycle upgrade. Expect future iterations to follow a similar pattern.