Logo
Overview
OpenAI Launches GPT-5.1: Adaptive Reasoning, 6 Personality Presets, and Group Chats After Mixed GPT-5 Reviews

OpenAI Launches GPT-5.1: Adaptive Reasoning, 6 Personality Presets, and Group Chats After Mixed GPT-5 Reviews

November 12, 2025
10 min read

On November 12, 2025, OpenAI launched GPT-5.1—a rapid iteration just three months after GPT-5’s August debut—introducing adaptive reasoning, a warmer conversational tone, six personality presets, and group chat functionality for up to 20 users. The quick turnaround comes after GPT-5 received mixed reviews from users, with complaints about its perceived coldness and rigidity. GPT-5.1 directly addresses this feedback with two variants: GPT-5.1 Instant (the default, now warmer and smarter) and GPT-5.1 Thinking (for complex reasoning tasks with dynamic compute allocation). Additionally, OpenAI is piloting group chats in ChatGPT in Japan, New Zealand, South Korea, and Taiwan—enabling collaborative AI-assisted conversations for teams. The update reflects OpenAI’s commitment to rapid iteration and user-centric design as competition heats up from Google Gemini, Anthropic Claude, and Meta’s Llama models.

Why GPT-5.1? Responding to User Feedback

GPT-5’s Rocky Reception

When OpenAI released GPT-5 in August 2025—more than two years after GPT-4—CEO Sam Altman called it “a significant step along our path to AGI”. The model delivered impressive benchmark improvements:

  • AIME 2025 (Advanced Math): 94.6% accuracy without tools
  • SWE-bench Verified (Coding): 74.9% success rate
  • MMMU (Multimodal Understanding): 84.2%
  • HealthBench Hard (Medical Reasoning): 46.2%

But users complained:

  • “GPT-5 feels robotic and cold compared to GPT-4”
  • “It’s too formal—I miss the conversational tone”
  • “GPT-5 doesn’t follow my instructions as well as GPT-4 did”
  • “Responses feel overly cautious and hedged”

The Numbers Backed This Up:

While OpenAI doesn’t release user satisfaction metrics, anecdotal reports on Reddit, Twitter, and developer forums suggested GPT-5 adoption lagged behind GPT-4 Turbo for everyday use cases—despite superior benchmarks. Users gravitated back to GPT-4 or switched to Claude 3.5 Sonnet, which was praised for its warmth and instruction-following.

OpenAI’s Response: GPT-5.1 in Record Time

Typically, OpenAI spaces major model releases 12-24 months apart:

  • GPT-3.5 (November 2022) → GPT-4 (March 2023): 4 months
  • GPT-4 (March 2023) → GPT-5 (August 2025): 29 months
  • GPT-5 (August 2025) → GPT-5.1 (November 2025): 3 months

The 3-month turnaround is unprecedented for OpenAI, signaling:

  1. User feedback was serious enough to warrant rapid intervention
  2. Competitive pressure from Claude 3.5 Sonnet and Gemini 2.0 Flash
  3. Iterative improvements don’t require full retraining—GPT-5.1 is likely a fine-tuned/aligned version of GPT-5, not a from-scratch rebuild

GPT-5.1 Instant: Warmer, Smarter, More Conversational

The Default Experience Reimagined

GPT-5.1 Instant replaces GPT-5 as the default model for ChatGPT Free, Go, Plus, and Pro users. Key improvements:

1. Adaptive Reasoning:

GPT-5.1 Instant can detect when a question is genuinely complex and allocate extra thinking time before responding—while answering simple questions instantly (no added latency).

How It Works:

  • Simple query (“What’s the capital of France?”): Immediate response (no thinking phase)
  • Complex query (“Design a distributed database schema for a multi-tenant SaaS app with GDPR compliance”): Model pauses to reason, then responds with thorough, accurate answer

The Technical Mechanism:

OpenAI trained GPT-5.1 with a meta-classifier that predicts question complexity. Based on this prediction:

  • Low complexity: Standard autoregressive generation (fast)
  • High complexity: Activates chain-of-thought reasoning (similar to GPT-5 Thinking, but shorter)

Benchmark Impact:

  • AIME 2025 (Math): Improves over GPT-5 Instant baseline
  • Codeforces (Competitive Programming): Stronger performance on hard problems
  • Latency: No degradation on simple queries—only adds time when needed

This is a major UX win: users get GPT-5 Thinking-level accuracy on hard problems without sacrificing GPT-4-level speed on easy ones.

2. Warmer by Default:

OpenAI retrained GPT-5.1 Instant to be “warmer and more conversational” compared to GPT-5. Internal testing showed:

  • Increased use of first-person language (“I think…” vs. “It could be argued that…”)
  • More natural phrasing—less formal, less hedged
  • Playfulness: Early testers report GPT-5.1 “surprises people with its playfulness while remaining clear and useful”

Example Comparison:

GPT-5 (Cold):

“The proposed approach could potentially yield benefits, although further analysis would be prudent before proceeding. One might consider alternative strategies as well.”

GPT-5.1 Instant (Warm):

“I think this approach looks promising! Here’s why it could work well… That said, you might also want to explore [alternative] if [condition]. What sounds best to you?”

3. Better Instruction Following:

Users complained GPT-5 didn’t follow instructions as reliably as GPT-4. GPT-5.1 addresses this with instruction-tuning improvements:

  • Tested on MixEval Hard (instruction-following benchmark): GPT-5.1 shows measurable improvement
  • Real-world examples: Users report GPT-5.1 sticks to format constraints (JSON output, word limits, tone requirements) more consistently

GPT-5.1 Thinking: Faster on Easy Tasks, Deeper on Hard Ones

The Reasoning Specialist

GPT-5.1 Thinking is OpenAI’s premium reasoning model, designed for:

  • Advanced mathematics (competition-level problems)
  • Complex coding (full-stack applications, algorithm design)
  • Research synthesis (academic literature review, hypothesis generation)
  • Strategic planning (business strategy, multi-step project planning)

Key Improvement: Dynamic Thinking Time

Unlike GPT-5 Thinking, which allocated fixed thinking time per query, GPT-5.1 Thinking adapts:

  • Easy task (“Explain quicksort”): ~2x faster than GPT-5 Thinking
  • Hard task (“Prove Fermat’s Last Theorem”): ~2x slower than GPT-5 Thinking (allocates more reasoning)

How This Works:

GPT-5.1 Thinking uses test-time compute scaling—the model dynamically decides how many reasoning steps to take before outputting a final answer. This is similar to:

  • AlphaGo’s variable-depth tree search
  • Google DeepMind’s AlphaProof (IMO gold medal winner)

Benchmark Performance:

OpenAI hasn’t released full GPT-5.1 Thinking benchmarks yet, but claims:

  • Competitive programming: Approaches Codeforces Expert level (top 5% of human programmers)
  • Math: Approaches AMC 12 (high school competition math) expert level

Six Personality Presets: Customize Your ChatGPT

The New Tone Controls

GPT-5.1 introduces six preset personalities users can select in ChatGPT settings:

  1. Default: Balanced, conversational, general-purpose
  2. Professional: Formal, concise, business-appropriate
  3. Friendly: Warm, encouraging, supportive
  4. Candid: Direct, no-nonsense, blunt when needed
  5. Quirky: Playful, creative, sometimes humorous
  6. Efficient: Minimal text, bullet points, no fluff

Legacy Presets (Still Available):

  • Nerdy: Technical jargon, deep dives, assumes expertise
  • Cynical: Skeptical, critical thinking, devil’s advocate

Granular Controls (Experimental)

Beyond presets, OpenAI is testing granular sliders in ChatGPT Labs:

  • Conciseness: Adjust response length (verbose ↔ terse)
  • Warmth: Formal ↔ friendly tone
  • Scannability: Dense paragraphs ↔ bullet points and headers
  • Emoji frequency: None ↔ frequent 🎉

Example Use Cases:

  • Developer: Nerdy + Efficient = terse, technical responses with code snippets
  • Content writer: Friendly + Quirky = creative, engaging copy
  • Executive: Professional + Concise = business-ready summaries

Competitive Context:

This feature directly competes with:

  • Anthropic Claude’s “tone” settings (Precise, Balanced, Creative)
  • Google Gemini’s customization options
  • Character.AI’s personality-driven chatbots

By offering 8 presets + granular controls, OpenAI positions ChatGPT as the most customizable mainstream AI assistant.

Group Chats: Collaborative AI for Teams

Pilot Launch in 4 Regions

OpenAI is piloting group chats in ChatGPT for users in:

  • 🇯🇵 Japan
  • 🇳🇿 New Zealand
  • 🇰🇷 South Korea
  • 🇹🇼 Taiwan

Key Features:

  • Up to 20 people in a single group chat
  • ChatGPT participates as a member—all participants see AI responses in real-time
  • Powered by GPT-5.1 Auto (dynamically selects best model based on query and user plan)
  • Available for: ChatGPT Free, Go, Plus, and Pro

Use Cases for Group Chats

1. Team Brainstorming:

  • Marketing team discusses campaign ideas
  • ChatGPT suggests creative concepts in real-time
  • Everyone iterates together

2. Coding Collaboration:

  • Developers debug a complex system
  • ChatGPT analyzes code snippets shared in chat
  • Provides suggestions visible to entire team

3. Education:

  • Study groups tackle homework problems
  • ChatGPT acts as a tutor, guiding without giving direct answers
  • Students learn collaboratively

4. Customer Support:

  • Support team handles complex ticket
  • ChatGPT drafts responses based on team discussion
  • Human agent reviews and sends

Why Pilot in These Regions?

Likely reasons:

  • Japan, South Korea, Taiwan: High ChatGPT adoption, tech-savvy users
  • New Zealand: English-speaking, smaller user base for manageable testing
  • Regulatory considerations: These regions have clearer AI regulations than EU (GDPR complexities) or China (restricted access)

Global Rollout Timeline:

OpenAI hasn’t announced dates, but typical pattern:

  • Pilot: 1-3 months (November 2025 - February 2026)
  • Gradual expansion: Add regions monthly
  • General availability: Mid-2026 (estimated)

Pricing and Availability

Who Gets GPT-5.1?

GPT-5.1 Instant:

  • ChatGPT Free: Limited access (20 messages every 3 hours, similar to current GPT-5 limits)
  • ChatGPT Plus ($20/month): Full access, higher rate limits
  • ChatGPT Pro ($200/month): Unlimited access, fastest response times
  • ChatGPT Go, Team, Business: Full access with team management features

GPT-5.1 Thinking:

  • ChatGPT Plus: 30 requests per day (estimated)
  • ChatGPT Pro: Unlimited thinking mode
  • API: Available via gpt-5.1-thinking endpoint (pricing TBD, likely similar to GPT-5 Thinking at $10-20 per million input tokens)

Group Chats:

  • All tiers in pilot regions (Free, Go, Plus, Pro)
  • Free tier likely has group size limits (e.g., max 5 people vs. 20 for paid)

Competitive Landscape: How GPT-5.1 Stacks Up

The AI Assistant Wars (November 2025)

ModelStrengthsWeaknessesPricing
GPT-5.1 InstantAdaptive reasoning, 6 personalities, group chatsStill behind Claude on instruction-following (per some users)$20/month (Plus)
Claude 3.5 SonnetBest instruction-following, warmest tone, 200K contextNo group chats, fewer integrations$20/month (Pro)
Gemini 2.0 FlashMultimodal output (images + audio), native tool use, fastestSmaller user base, fewer pluginsFree (limited), $20/month (Advanced)
Meta Llama 4 405BOpen-source, self-hostable, no data sharingRequires technical setup, no official UIFree (self-hosted)

GPT-5.1’s Positioning:

  • Wins on: Customization (personality presets), collaboration (group chats), ecosystem (plugins, integrations)
  • Competitive on: Reasoning (adaptive thinking), speed (Instant mode)
  • Still trails on: Warmth/instruction-following (Claude), multimodal output (Gemini)

The Broader Strategy: Rapid Iteration Beats Perfection

OpenAI’s New Playbook

GPT-5.1’s 3-month turnaround signals a strategic shift:

Old Strategy (2022-2024):

  • Train massive model for 12-24 months
  • Launch with big fanfare
  • Wait for next generation before addressing issues

New Strategy (2025+):

  • Launch “good enough” model (GPT-5)
  • Listen to users
  • Rapidly iterate with fine-tuned versions (GPT-5.1, GPT-5.2…)
  • Compete with Claude’s frequent updates and Gemini’s experimental models

Why the Change?

  1. Competition intensified: Claude 3.5 Sonnet and Gemini 2.0 Flash launched months after GPT-5—OpenAI couldn’t wait 2 years to respond
  2. User expectations evolved: Developers expect continuous improvement, not annual releases
  3. Fine-tuning is faster: GPT-5.1 likely didn’t require full retraining—mostly RLHF (Reinforcement Learning from Human Feedback) adjustments

Conclusion: OpenAI Listens and Delivers

GPT-5.1 demonstrates OpenAI’s responsiveness to user feedback and commitment to rapid iteration. By addressing GPT-5’s coldness and rigidity with a warmer tone, adaptive reasoning, and personality customization—while piloting group chats for collaborative use—OpenAI keeps ChatGPT competitive against Claude, Gemini, and open-source alternatives.

Key takeaways:

  1. Adaptive reasoning bridges the gap between speed (Instant) and depth (Thinking)
  2. Personality presets make ChatGPT the most customizable AI assistant
  3. Group chats unlock team collaboration use cases (education, business, development)
  4. 3-month iteration cycle is OpenAI’s answer to Claude and Gemini’s frequent updates

As the AI assistant market matures, differentiation comes from UX polish—not just raw capabilities. GPT-5.1 shows OpenAI understands this: users don’t just want smarter AI, they want AI that feels right. With warmer tone, flexible personalities, and collaborative features, GPT-5.1 makes ChatGPT feel less like a tool and more like a teammate.

The question now: Will users return from Claude and Gemini—or has OpenAI already lost mindshare during GPT-5’s rocky months?