OpenAI released GPT-5.1 on November 12, 2025, promoting it as an evolutionary update that blends faster responses with more flexible conversational style. The rollout began the same day for paid tiers , Pro, Plus, Go, and Business , with gradual availability to free and logged-out users and a seven-day enterprise early-access toggle for companies that want to test before enabling it organization-wide.
Beyond the release schedule, the line features are twofold: a new adaptive thinking core capability that dynamically varies reasoning effort, and an expanded set of personality presets and personalization experiments to restore warmth and give users more control. The update responds to earlier user feedback about tone while promising measurable latency and accuracy gains for developers and partners.
What GPT-5.1 is and how it differs
GPT-5.1 ships in two primary variants: GPT-5.1 Instant, an instruction-following default optimized for warm, fast replies, and GPT-5.1 Thinking, which is tuned for deeper, persistent reasoning on complex tasks. In most cases queries are auto-matched to the appropriate variant so that users get instant answers when appropriate and more sustained thinking for harder problems.
The launch packaging also keeps legacy access practical: GPT-5 remains available in ChatGPTs legacy models dropdown for three months after the GPT-5.1 launch so users and teams can compare behavior and migrate at their own pace. OpenAI framed the change as an incremental upgrade rather than a forced replacement.
Context matters: the update follows a period of community pushback when some users found GPT-5 colder and less personable. Company leaders signaled a move to restore warmth and offer finer-grained customization, a rationale reflected in this release's emphasis on conversational style alongside capability improvements.
Adaptive thinking: dynamic reasoning effort
The core technical line for GPT-5.1 is adaptive reasoning, sometimes called adaptive thinking. The model dynamically varies how much time and how many tokens it spends 'thinking' based on task complexity. Simple lookups and routine edits return quickly; multi-step planning or novel problems trigger longer, more persistent internal effort.
For developers, OpenAI exposed control points such as reasoning_effort, including a no-reasoning mode where reasoning_effort='none' optimizes for low latency. In partner evaluations that no-reasoning mode produced about a 20% improvement in low-latency tool-calling vs GPT-5 minimal reasoning, a useful tradeoff for interactive systems and live agents.
OpenAI provided an example latency improvement: answering an npm list command in roughly 2 seconds (about 50 tokens) with GPT-5.1 versus roughly 10 seconds (about 250 tokens) for GPT-5 in their comparison. That kind of jump demonstrates how adaptive thinking can reduce wasted compute on simple tasks while still offering deeper reasoning when needed.
Performance, partner reports, and benchmarks
Partners reported strong results. Investment firm Balyasny Asset Management said GPT-5.1 outperformed both GPT-4.1 and GPT-5 while running two to three times faster than GPT-5 in their workloads. Pace reported agents running 50% faster while also exceeding GPT-5 accuracy in their internal tests. OpenAI cites these partner quotes when describing real-world gains.
Benchmarks in OpenAI's appendix show measurable accuracy improvements on several public tests: for example, SWE-bench Verified rose to 76.3% for GPT-5.1 compared with 72.8% for GPT-5, and GPQA Diamond reached 88.1% versus 85.7%. These are modest but meaningful improvements that align with both lower latency and higher correctness.
Taken together, the partner anecdotes and benchmark lifts suggest GPT-5.1 aims to be both faster and smarter across common developer and enterprise tasks, not merely a cosmetic update to style or tone.
Developer features, APIs, and new tools
GPT-5.1 is available in the API under several model names: gpt-5.1 for Thinking and gpt-5.1-chat-latest for Instant. Pricing and rate limits remain consistent with GPT-5, and OpenAI also added gpt-5.1-codex and gpt-5.1-codex-mini models for coding tasks.
Two new developer tools debuted with the release. The freeform apply_patch tool lets the model propose structured code edits, simplifying code review and iteration. The shell tool allows the model to propose shell commands which a developer runs and then returns output to the model, tightening the loop between suggestions and results while preserving human control.
Operational improvements include extended prompt caching with up to 24-hour retention to speed follow-ups and reduce costs. OpenAI says cached input tokens are about 90% cheaper than uncached tokens, a potentially significant cost saver for high-volume conversational apps and agents that reuse recent context.
Personality presets and personalization experiments
One of the most visible consumer-facing changes is the expanded set of personality presets in ChatGPT. The presets now include Default, Friendly, Efficient, Professional, Candid, Quirky, plus the retained Nerdy and Cynical options , eight total. These presets apply across models so users can choose a tone that fits their task or mood.
OpenAI is also experimenting with more granular personalization controls: sliders for concision, warmth, scannability, and emoji frequency, and in-conversation proactive suggestions to update style as the chat evolves. Those experiments are rolling out gradually to a limited set of users as part of a staged approach to personalization.
The move toward richer personality controls follows public feedback about GPT-5's colder tone; the company has framed the presets as a way to restore warmth while allowing enterprise and power users to keep more task-focused behavior. With more than 800 million people using ChatGPT globally, OpenAI cites scale as the reason finer personalization matters.
Rollout, policy, and industry response
The rollout strategy balanced speed and control: paid users got access immediately on November 12, 2025, with gradual expansion to free users and a seven-day early-access toggle for enterprise customers that want to test internally before flipping GPT-5.1 on for everyone. This staged approach aims to reduce disruptions for large organizations.
OpenAI also announced a three-month availability window for GPT-5 in the legacy dropdown so teams can compare models before deprecation. That policy gives developers and creators breathing room to validate agents, integrations, and user experiences against the new behavior and performance characteristics.
Press coverage framed GPT-5.1 as an 'upgrade' that redresses concerns about tone while preserving frontier capabilities. Industry commentary highlights a deliberate attempt to balance conversational warmth and personality customization with improved reasoning efficiency and developer ergonomics.
GPT-5.1 combines adaptive reasoning and richer personality controls to address both technical and user-experience critiques of the prior release. For engineers and enterprises the line is better latency, new developer tools, and modest benchmark gains; for end users it is more choices about how their assistant sounds and behaves.
Whether you prioritize speed, deeper reasoning, or a specific conversational voice, GPT-5.1 aims to provide configurable tradeoffs. As the model reaches wider availability, the three-month legacy window and ongoing personalization experiments should give users and organizations time to find the settings that work best for them.
