GPT-5.1

2025-11-12 · Publisher : OpenAI · Domain : Artificial Intelligence

OpenAI launches GPT-5.1 with two models in ChatGPT, GPT-5.1 Instant and GPT-5.1 Thinking, warmer and more reliable, then offers them in the API the next day with adaptive reasoning.

What changes

  • GPT-5.1 Instant, the most used ChatGPT model, more conversational and better at following instructions; GPT-5.1 Thinking faster on simple tasks and more persistent on difficult ones.
  • New tone and style settings in ChatGPT.
  • API (November 13): gpt-5.1 (Thinking) and gpt-5.1-chat-latest (Instant), with adaptive reasoning and a no-reasoning mode (reasoning_effort “none”, the default).
  • Prompt cache extended up to 24 hours, apply_patch and shell tools, and gpt-5.1-codex and gpt-5.1-codex-mini models.
  • GPT-5 remains available for three months among the legacy models in ChatGPT.

### Specifications - Context 400,000 tokens, output 128,000 tokens, text and image input, text output, knowledge cutoff September 30, 2024. - API identifier: gpt-5.1 (snapshot gpt-5.1-2025-11-13); effort levels none, low, medium, high. - ChatGPT rollout: paid subscribers first, then free and logged-out users; Enterprise and Edu with seven days of early access. ### Pricing Same prices as GPT-5, per million tokens (USD), with cache 90% cheaper. | | Input | Cached input | Output | | --- | --- | --- | --- | | gpt-5.1 | 1.25 | 0.125 | 10 | | gpt-5.1-codex-mini | 0.25 | 0.025 | 2 | ### Benchmarks Scores published by the vendor (high effort), GPT-5.1 versus GPT-5. | Benchmark | GPT-5.1 | GPT-5 | | --- | --- | --- | | SWE-bench Verified (500 tasks) | 76.3% | 72.8% | | GPQA Diamond | 88.1% | 85.7% | | AIME 2025 | 94.0% | 94.6% | | MMMU | 85.4% | 84.2% | | Tau2-bench Telecom | 95.6% | 96.7% | ### Key takeaways GPT-5.1 mainly improves latency and style rather than raw intelligence. The no-reasoning mode and the 24-hour cache target high-volume interactive applications.

Links

Tags : LLM, OpenAI, GPT, GPT-5.1, Raisonnement adaptatif, Agent