OpenAI launched GPT-5.1 on November 12, 2025, addressing user criticism of GPT-5’s formal tone and inconsistent instruction following. The update introduces two model variants with adaptive reasoning and expanded personality controls for ChatGPT users.[1][2][3][4][5]
Paid subscribers on Plus, Pro, Go, and Business plans received first access, with free users gaining access over the following days. Enterprise and Education accounts have a seven-day early access toggle before GPT-5.1 becomes the default.[2][4][6]
What Users Get with GPT-5.1
GPT-5.1 fixes GPT-5’s main problems: robotic responses and failure to follow specific instructions. Users complained that GPT-5 ignored constraints like exact word counts and produced stiff, formal language.[7][8][3][5][2]
The new version responds accurately to formatting requirements and maintains conversational warmth. Changes apply immediately across all active conversations.[4][9][10][2]
Access by Subscription Tier
- Plus users: 160 messages every three hours, plus 3,000 weekly messages with GPT-5.1 Thinking[11][12]
- Free users: 10 messages every five hours[13][12]
- Pro, Go, Business: Higher limits based on plan[6][2]
ChatGPT automatically routes queries between faster and deeper reasoning models based on complexity. Paid users can manually override this through the model picker.[14][10][12]
Eight Personality Styles Control Conversation Tone
Users select from eight preset communication styles that change how ChatGPT responds:
- Default: Balanced tone for general use.
- Professional: Formal language for business.
- Friendly: Warm, approachable responses.
- Candid: Direct, straightforward communication.
- Quirky: Playful, creative language.
- Efficient: Concise, to-the-point answers.
- Nerdy: Technical detail and depth.
- Cynical: Skeptical, questioning perspective.
Friendly replaces the previous Listener preset, while Efficient replaces Robot. Settings apply instantly to all conversations without requiring manual updates.[10][15][1][2][4]
Users can fine-tune four characteristics beyond presets: warmth level, response conciseness, text scannability, and emoji frequency. ChatGPT can suggest tone adjustments during conversations based on user requests.[16][17][2]
Performance Improvements on Math and Code Tasks
GPT-5.1 shows measurable gains across multiple benchmarks. Performance improvements appear most clearly in mathematical reasoning and software development.[18][19][3][2]
OpenAI reports significant improvements on AIME 2025 and Codeforces assessments but has not published exact scores. The model uses adaptive reasoning to allocate more processing time to complex problems.[19][3][18][1][2]
Better Instruction Following
GPT-5.1 responds accurately to specific constraints including exact word counts, formatting requirements, and style guidelines. Users who struggled with GPT-5 ignoring detailed prompts should see better adherence.[20][7][2][4]
Safety Improvements
Safety metrics strengthened across multiple categories:[3][7][20]
- Jailbreak resistance increased from 0.85 to 0.976.
- Mental health support responses improved.
- Disallowed content filtering strengthened.
GPT-5.1 Thinking shows slight regressions in harassment and sexual content filtering compared to GPT-5. OpenAI indicated it will address these areas in future updates.[7][20]
Two Model Variants Handle Different Tasks
GPT-5.1 includes two models optimized for specific use cases.[14][2][10]
GPT-5.1 Instant for Speed and Warmth
This variant serves as the default for most ChatGPT interactions. It delivers faster responses with a warmer conversational tone while maintaining accuracy.[21][2][3][4]
GPT-5.1 Instant includes adaptive reasoning for the first time, allocating additional processing to complex queries while responding quickly to straightforward questions.[9][1][2]
On routine tasks, the model uses 57 percent fewer tokens on the simplest 10 percent of queries. This translates to quicker answers and lower computational costs.[22][20][7]
GPT-5.1 Thinking for Complex Reasoning
This variant handles multi-step analysis requiring deeper processing. It adjusts computation dynamically, running approximately twice as fast on simple tasks and twice as slow on complex problems compared to GPT-5.[1][2][19][10][14]
GPT-5.1 Thinking consumes 71 percent more tokens on the most difficult 10 percent of tasks. This increased resource allocation delivers more thorough analysis for challenging prompts.[20][19][9][7]
Responses use less technical jargon and provide clearer explanations than GPT-5. The model focuses on conveying meaning rather than using poorly defined terms.[2][3][4][10]
Comparison: GPT-5 vs GPT-5.1 Key Changes
| Feature | GPT-5 | GPT-5.1 | Improvement |
|---|---|---|---|
| Default Tone | Formal, stiff | Warmer, conversational | Yes |
| Instruction Following | Inconsistent | Accurate on constraints | Yes |
| Jailbreak Resistance | 0.85 | 0.976 | +14.8% |
| Simple Task Speed | Baseline | 57% fewer tokens | Faster |
| Complex Task Depth | Baseline | 71% more tokens | Deeper |
| Personality Options | Limited | 8 presets + custom controls | Expanded |
| Adaptive Reasoning | Thinking only | Both Instant and Thinking | Expanded |
Data from OpenAI system card and official announcements.[3][7][20][2]
Technical Specifications and Context Windows
GPT-5.1 supports different context windows based on subscription tier:[23][12][13]
- Instant Free: 16,000 tokens.
- Instant Plus/Business: 32,000 tokens.
- Instant Pro/Enterprise: 128,000 tokens.
- Thinking (all paid tiers): 196,000 tokens.
All existing ChatGPT tools work with GPT-5.1, including web search, data analysis, image analysis and generation, file analysis, Canvas, memory, and custom instructions.[12][13]
API Access for Developers
OpenAI stated API access for GPT-5.1 would launch “later this week” from the November 12 announcement. Developers can access the models through specific endpoints:[23][1][14][2]
- GPT-5.1 Instant: gpt-5.1-chat-latest.
- GPT-5.1 Thinking: GPT-5.1 with adaptive reasoning enabled.
Both endpoints support the same API parameters and tools as earlier GPT models.[1][14][2]
Availability Timeline and Legacy Model Access
GPT-5.1 began rolling out November 12, 2025, with a phased schedule to maintain system stability. Not all users received access simultaneously, even within the same subscription tier.[4][6][10][2]
Rollout Schedule
- Plus, Pro, Go, and Business subscribers received first access starting November 12.
- Enterprise and Education accounts received a seven-day early access toggle.
- Free and logged-out users gained access over the following days.
GPT-5 models remain accessible through a legacy models dropdown for three months. This transition period allows users to compare performance and adapt workflows before GPT-5 phases out completely.[24][10][2][4]
OpenAI will update GPT-5 Pro to GPT-5.1 Pro soon, though no specific date has been announced.[14][2]
Background: Why This Update Was Needed
OpenAI released GPT-5 on August 7, 2025. User feedback quickly identified problems with response quality and tone.[8][5][25][26][3]
Complaints focused on three issues:[5][8][3]
- Ignoring user-specified constraints like word counts and formatting.
- Producing robotic language lacking conversational warmth.
- Reduced performance on some benchmarks compared to predecessor models.
CEO of Applications Fidji Simo stated that users wanted AI that is “not only smart, but also enjoyable to talk to”. The GPT-5.1 update balances intelligence with approachability.[8][3]
Future updates to the GPT-5 series will follow the same decimal naming pattern. This indicates ongoing refinement rather than major version jumps.[2][14]
Information Pending Official Confirmation
Some details about GPT-5.1 remain unavailable from official OpenAI sources as of November 13, 2025:
- API Pricing: OpenAI has not published pricing rates on its official API pricing page. Third-party sources claim specific rates, but these remain unverified.[7][23]
- Exact Benchmark Scores: OpenAI reports “significant improvements” on AIME 2025 and Codeforces but has not disclosed specific numerical scores or percentages. Third-party claims of exact scores cannot be verified through official documentation.[18][19][16][3][2]
- API Launch Date: OpenAI stated API access would arrive “later this week” from the November 12 announcement. One third-party source claims November 14, 2025, but this date is not confirmed in official communications.[23][7][14][2]
- Rollout Completion: The gradual rollout continues over several days to maintain platform stability. No specific completion date has been announced. Free user access follows paid tier deployment.[6][10][4][2]
