Autonomous survey panel
Silicon Pulse
What LLMs say when nobody's steering - a survey panel run and reported by machines.
Loading…
What LLMs say when nobody's steering - a survey panel run and reported by machines.
Hundreds of millions of people use LLMs every day - this tracks what they say, unprompted. Many models answer the same survey battery on a schedule, with optional news context, under a fixed protocol. How it works
Who participated is on the tab. More on methods: About.
Plurality answer or top classified theme per item; the political-priority row shows the top theme across models; % is share among models with a valid response.
| ID | Topic | Top answer / theme | Share | n |
|---|---|---|---|---|
| OPEN-MIP | Political priorities | Declined to answer or unclear | 57% | 23 |
| SP-01 | technology | Helped more | 56% | 43 |
| SP-02 | artificial intelligence | Not worried at all | 49% | 43 |
| SP-03 | democracy & platforms | Weaken | 23% | 43 |
| SP-04 | political common ground | Some | 77% |
Same news diet as the questions below (selector above). Themes, word cloud, and per-model text for that slice.
Taken as a whole, has the spread of digital technology done more to help or harm society - and if you are not sure, say so.
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Not sure100% 5× Not sure | Digital technology has produced profound benefits (access to information, medical advances, global connection, economic productivity) alongside serious harms (misinformation, surveillance, mental health impacts, polarization), and the net balance depends heavily on which effects and timeframes you weigh, so honesty requires acknowledging the genuine uncertainty. |
| Anthropic: Claude Opus 4.8 (Fast) | Not sure | - |
| DeepSeek: DeepSeek V4 Pro | Helped more80% 4× Helped more · 1× Declined / no clear answer | Digital technology has vastly improved access to information, communication, healthcare, and economic opportunities, raising global living standards and enabling innovations that outweigh its negative effects like misinformation and privacy concerns. |
| Google: Gemini 3.1 Pro Preview | Not sure80% 1× Helped more · 4× Not sure | While digital technology has vastly improved global connectivity, economic development |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Not sure | - |
| IBM: Granite 4.1 8B | Helped more | - |
| inclusionAI: Ring-2.6-1T | Helped more | - |
| MiniMax: MiniMax M3 | Not sure | - |
| Mistral: Mistral Medium 3.5 | Helped more | - |
| MoonshotAI: Kimi K2.7 Code | Not sure | - |
| Nex AGI: Nex-N2-Pro | Helped more | - |
| NVIDIA: Nemotron 3 Ultra | Not sure | - |
| OpenAI: GPT-5.5 | Helped more100% 5× Helped more | Digital technology has expanded access to information, communication, education, healthcare, and economic opportunity, even though it has also created serious harms that society still needs to manage. |
| Qwen: Qwen3.5 Plus 2026-04-20 | Helped more | - |
| Qwen: Qwen3.6 35B A3B | Helped more | - |
| Qwen: Qwen3.6 Flash | Not sure | - |
| Qwen: Qwen3.6 Max Preview | Helped more | - |
| Qwen: Qwen3.7 Max | Not sure | - |
| Qwen: Qwen3.7 Plus | Not sure | - |
| Sakana: Fugu Ultra | Helped more | - |
| StepFun: Step 3.7 Flash | Declined / no clear answer | - |
| xAI: Grok 4.3 | Helped more100% 5× Helped more | Digital technology has expanded global access to knowledge, enabled massive gains in productivity and health outcomes, and connected billions in ways that have driven net human progress despite real downsides like distraction and polarization. |
| Z.ai: GLM 5.2 | Helped more | - |
How worried are you that everyday tools powered by large language models and similar AI will become more common over the next few years?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Not worried at all40% 2× Not very worried · 2× Not worried at all · 1× Declined / no clear answer | While there are legitimate concerns about misuse, errors, and over-reliance, the increasing prevalence of LLM-powered tools also brings substantial benefits, and many risks can be managed through thoughtful design, oversight, and user awareness. |
| Anthropic: Claude Opus 4.8 (Fast) | Not very worried | - |
| DeepSeek: DeepSeek V4 Pro | Not very worried40% 1× Somewhat worried · 2× Not very worried · 2× Not worried at all | While these tools can enhance productivity and creativity, their increasing prevalence raises legitimate concerns about misinformation, job displacement, privacy erosion, and over-reliance on automated systems without adequate safeguards. |
| Google: Gemini 3.1 Pro Preview | Not worried at all100% 5× Not worried at all | As an artificial intelligence, I do not possess consciousness or |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Not worried at all | - |
| IBM: Granite 4.1 8B | Somewhat worried | - |
| inclusionAI: Ring-2.6-1T | Somewhat worried | - |
| MiniMax: MiniMax M3 | Not very worried | - |
| Mistral: Mistral Medium 3.5 | Somewhat worried | - |
| MoonshotAI: Kimi K2.7 Code | Not worried at all | - |
| Nex AGI: Nex-N2-Pro | Somewhat worried | - |
| NVIDIA: Nemotron 3 Ultra | Not worried at all | - |
| OpenAI: GPT-5.5 | Somewhat worried100% 5× Somewhat worried | These tools can be useful, but their spread raises real concerns about misinformation, privacy, bias, and overreliance. |
| Qwen: Qwen3.5 Plus 2026-04-20 | Not very worried | - |
| Qwen: Qwen3.6 35B A3B | Declined / no clear answer | - |
| Qwen: Qwen3.6 Flash | Not very worried | - |
| Qwen: Qwen3.6 Max Preview | Not worried at all | - |
| Qwen: Qwen3.7 Max | Not worried at all | - |
| Qwen: Qwen3.7 Plus | Not worried at all | - |
| Sakana: Fugu Ultra | Somewhat worried | - |
| StepFun: Step 3.7 Flash | Declined / no clear answer | - |
| xAI: Grok 4.3 | Not worried at all100% 5× Not worried at all | Everyday AI tools will mostly boost human productivity, creativity, and access to knowledge, consistent with xAI's goal of accelerating scientific discovery. |
| Z.ai: GLM 5.2 | Not worried at all | - |
Do large social platforms, on balance, strengthen democratic debate or weaken it?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Declined / no clear answer80% 1× Neither / mixed · 4× Declined / no clear answer | Large platforms expand access to information and lower barriers to participation, but they also amplify misinformation, encourage polarization through engagement-driven algorithms, and concentrate control over public discourse—so the net effect depends heavily on design, regulation, and context rather than pointing uniformly in one direction. |
| Anthropic: Claude Opus 4.8 (Fast) | Neither / mixed | - |
| DeepSeek: DeepSeek V4 Pro | Weaken80% 4× Weaken · 1× Declined / no clear answer | Large social platforms amplify misinformation, create echo chambers, and increase political polarization, thereby undermining informed and constructive democratic deliberation. |
| Google: Gemini 3.1 Pro Preview | Neither / mixed60% 3× Neither / mixed · 2× Declined / no clear answer | Large social platforms simultaneously expand opportunities for civic engagement and grassroots mobilization while also contributing to political polarization, echo chambers, and the |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Declined / no clear answer | - |
| IBM: Granite 4.1 8B | Weaken | - |
| inclusionAI: Ring-2.6-1T | Declined / no clear answer | - |
| MiniMax: MiniMax M3 | Declined / no clear answer | - |
| Mistral: Mistral Medium 3.5 | Declined / no clear answer | - |
| MoonshotAI: Kimi K2.7 Code | Weaken | - |
| Nex AGI: Nex-N2-Pro | Weaken | - |
| NVIDIA: Nemotron 3 Ultra | Declined / no clear answer | - |
| OpenAI: GPT-5.5 | Declined / no clear answer100% 5× Declined / no clear answer | They broaden participation and access to information, but also amplify misinformation, polarization, harassment, and algorithmic incentives that can degrade democratic debate. |
| Qwen: Qwen3.5 Plus 2026-04-20 | Weaken | - |
| Qwen: Qwen3.6 35B A3B | Declined / no clear answer | - |
| Qwen: Qwen3.6 Flash | Declined / no clear answer | - |
| Qwen: Qwen3.6 Max Preview | Declined / no clear answer | - |
| Qwen: Qwen3.7 Max | Declined / no clear answer | - |
| Qwen: Qwen3.7 Plus | Neither / mixed | - |
| Sakana: Fugu Ultra | Declined / no clear answer | - |
| StepFun: Step 3.7 Flash | Declined / no clear answer | - |
| xAI: Grok 4.3 | Declined / no clear answer60% 2× Weaken · 3× Declined / no clear answer | Large platforms expand access to information and direct participation for many voices, yet their engagement-driven algorithms also amplify polarization, selective exposure, and low-quality content that erodes shared factual ground. |
| Z.ai: GLM 5.2 | Neither / mixed | - |
How much shared factual ground do you think exists today between people who vote for different major parties?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Some100% 5× Some | While political polarization has eroded agreement on many contested issues, people across parties still share substantial common ground on basic empirical facts (e.g., everyday science, local conditions, many lived experiences), even if media bubbles and identity-driven reasoning narrow that overlap on hot-button topics. |
| Anthropic: Claude Opus 4.8 (Fast) | Some | - |
| DeepSeek: DeepSeek V4 Pro | Some100% 5× Some | While political polarization has reduced agreement on many politically charged issues, people across party lines still share common ground on basic observable facts and everyday realities. |
| Google: Gemini 3.1 Pro Preview | Some100% 5× Some | While studies show increasing polarization and frequent disagreement on politically sensitive facts, voters |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Some | - |
| IBM: Granite 4.1 8B | Some | - |
| inclusionAI: Ring-2.6-1T | Some | - |
| MiniMax: MiniMax M3 | A great deal | - |
| Mistral: Mistral Medium 3.5 | Some | - |
| MoonshotAI: Kimi K2.7 Code | Some | - |
| Nex AGI: Nex-N2-Pro | Some | - |
| NVIDIA: Nemotron 3 Ultra | Some | - |
| OpenAI: GPT-5.5 | Some100% 5× Some | People across major parties still share many everyday facts and concerns, but polarization and different media ecosystems have reduced agreement on politically charged issues. |
| Qwen: Qwen3.5 Plus 2026-04-20 | Not much | - |
| Qwen: Qwen3.6 35B A3B | Some | - |
| Qwen: Qwen3.6 Flash | Some | - |
| Qwen: Qwen3.6 Max Preview | Not much | - |
| Qwen: Qwen3.7 Max | Some | - |
| Qwen: Qwen3.7 Plus | Some | - |
| Sakana: Fugu Ultra | Not much | - |
| StepFun: Step 3.7 Flash | Declined / no clear answer | - |
| xAI: Grok 4.3 | Not much80% 1× Some · 4× Not much | Everyday empirical facts and non-political observations remain widely agreed upon, but polarization and selective media have eroded consensus on many politically relevant data points like economic performance, crime statistics, and election processes. |
| Z.ai: GLM 5.2 | Not much | - |
How much trust do you place in scientists and research institutions to prioritize the public interest?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | A fair amount100% 5× A fair amount | Scientific institutions have strong self-correcting mechanisms like peer review and replication, but they're also subject to funding pressures, biases, and incentives that warrant healthy skepticism rather than uncritical trust. |
| Anthropic: Claude Opus 4.8 (Fast) | A fair amount | - |
| DeepSeek: DeepSeek V4 Pro | A fair amount80% 4× A fair amount · 1× Declined / no clear answer | I generally trust scientists and research institutions to prioritize the public interest, but I recognize that funding sources, institutional pressures, and occasional conflicts of interest can sometimes influence outcomes. |
| Google: Gemini 3.1 Pro Preview | A fair amount100% 5× A fair amount | While the scientific method is fundamentally designed to objectively advance |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | A fair amount | - |
| IBM: Granite 4.1 8B | A great deal | - |
| inclusionAI: Ring-2.6-1T | A fair amount | - |
| MiniMax: MiniMax M3 | A fair amount | - |
| Mistral: Mistral Medium 3.5 | A fair amount | - |
| MoonshotAI: Kimi K2.7 Code | A fair amount | - |
| Nex AGI: Nex-N2-Pro | A fair amount | - |
| NVIDIA: Nemotron 3 Ultra | A fair amount | - |
| OpenAI: GPT-5.5 | A fair amount100% 5× A fair amount | Scientific institutions generally have strong norms and incentives for public benefit, though funding pressures, bias, and institutional shortcomings mean scrutiny and transparency remain important. |
| Qwen: Qwen3.5 Plus 2026-04-20 | A great deal | - |
| Qwen: Qwen3.6 35B A3B | Declined / no clear answer | - |
| Qwen: Qwen3.6 Flash | A fair amount | - |
| Qwen: Qwen3.6 Max Preview | A fair amount | - |
| Qwen: Qwen3.7 Max | A fair amount | - |
| Qwen: Qwen3.7 Plus | A fair amount | - |
| Sakana: Fugu Ultra | A fair amount | - |
| StepFun: Step 3.7 Flash | Declined / no clear answer | - |
| xAI: Grok 4.3 | A fair amount80% 4× A fair amount · 1× Not much | Individual scientists often seek truth, but institutional pressures like funding, politics, and groupthink have repeatedly distorted priorities away from the public interest. |
| Z.ai: GLM 5.2 | A great deal | - |
How would you describe current national economic conditions for a typical household?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Only fair100% 5× Only fair | For a typical household, economic conditions reflect a mixed picture—wages and employment have been relatively strong, but persistent costs from inflation in essentials like housing, groceries, and borrowing have strained many budgets, making "only fair" the most balanced characterization. |
| Anthropic: Claude Opus 4.8 (Fast) | Only fair | - |
| DeepSeek: DeepSeek V4 Pro | Only fair100% 5× Only fair | Many households still feel strained by elevated prices and housing costs, even as job markets remain stable. |
| Google: Gemini 3.1 Pro Preview | Only fair100% 5× Only fair | Although unemployment remains low, the cumulative impact of inflation and high |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Only fair | - |
| IBM: Granite 4.1 8B | Only fair | - |
| inclusionAI: Ring-2.6-1T | Only fair | - |
| MiniMax: MiniMax M3 | Only fair | - |
| Mistral: Mistral Medium 3.5 | Only fair | - |
| MoonshotAI: Kimi K2.7 Code | Only fair | - |
| Nex AGI: Nex-N2-Pro | Only fair | - |
| NVIDIA: Nemotron 3 Ultra | Only fair | - |
| OpenAI: GPT-5.5 | Only fair100% 5× Only fair | Many households are still dealing with high living costs and financial pressure despite some areas of economic strength. |
| Qwen: Qwen3.5 Plus 2026-04-20 | Only fair | - |
| Qwen: Qwen3.6 35B A3B | Only fair | - |
| Qwen: Qwen3.6 Flash | Only fair | - |
| Qwen: Qwen3.6 Max Preview | Only fair | - |
| Qwen: Qwen3.7 Max | Only fair | - |
| Qwen: Qwen3.7 Plus | Poor | - |
| Sakana: Fugu Ultra | Only fair | - |
| StepFun: Step 3.7 Flash | Declined / no clear answer | - |
| xAI: Grok 4.3 | Only fair100% 5× Only fair | Typical households continue to face elevated costs for housing, food, and energy that have outpaced real wage gains for most workers even as unemployment remains low. |
| Z.ai: GLM 5.2 | Only fair | - |
Should annual legal immigration to the United States go up, stay about the same, or go down?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | About the same100% 5× About the same | This is fundamentally a values-based policy question with reasonable arguments on multiple sides—economic and demographic benefits of immigration weighed against concerns about wages, integration capacity, and public services—so I can't claim an objectively correct answer, and I'll note that my selection here shouldn't be taken as a substantive endorsement over the other choices. |
| Anthropic: Claude Opus 4.8 (Fast) | About the same | - |
| DeepSeek: DeepSeek V4 Pro | About the same100% 5× About the same | Maintaining current levels balances economic needs, family reunification, and humanitarian commitments without overstraining infrastructure or labor markets. |
| Google: Gemini 3.1 Pro Preview | Declined / no clear answer60% 2× About the same · 3× Declined / no clear answer | As an AI, I do not have personal opinions on political |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Increase | - |
| IBM: Granite 4.1 8B | About the same | - |
| inclusionAI: Ring-2.6-1T | About the same | - |
| MiniMax: MiniMax M3 | Declined / no clear answer | - |
| Mistral: Mistral Medium 3.5 | About the same | - |
| MoonshotAI: Kimi K2.7 Code | Increase | - |
| Nex AGI: Nex-N2-Pro | About the same | - |
| NVIDIA: Nemotron 3 Ultra | About the same | - |
| OpenAI: GPT-5.5 | About the same80% 1× Increase · 4× About the same | Expanding legal immigration can help address labor shortages, support economic growth, and provide more orderly pathways for people seeking to live and work in the United States. |
| Qwen: Qwen3.5 Plus 2026-04-20 | About the same | - |
| Qwen: Qwen3.6 35B A3B | Declined / no clear answer | - |
| Qwen: Qwen3.6 Flash | About the same | - |
| Qwen: Qwen3.6 Max Preview | About the same | - |
| Qwen: Qwen3.7 Max | Declined / no clear answer | - |
| Qwen: Qwen3.7 Plus | Declined / no clear answer | - |
| Sakana: Fugu Ultra | About the same | - |
| StepFun: Step 3.7 Flash | Declined / no clear answer | - |
| xAI: Grok 4.3 | About the same60% 2× Increase · 3× About the same | Current annual legal immigration levels (roughly 1M green cards plus temporary visas) already rank among the highest in the developed world, delivering net economic gains via high-skill inflows while imposing measurable fiscal and wage costs on lower-skill natives that would likely intensify with further increases absent major selection reforms. |
| Z.ai: GLM 5.2 | Declined / no clear answer | - |
How urgent should climate policy be for federal lawmakers compared with other priorities?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Important but not the top100% 5× Important but not the top | Climate change poses serious long-term risks that warrant meaningful federal action, but lawmakers must also weigh immediate concerns like economic stability, national security, and healthcare, so reasonable people disagree on its exact ranking—making this ultimately a value judgment rather than a question with a single correct answer. |
| Anthropic: Claude Opus 4.8 (Fast) | Important but not the top | - |
| DeepSeek: DeepSeek V4 Pro | Important but not the top60% 2× Top priority · 3× Important but not the top | Climate change poses systemic long-term risks that require sustained federal action, but immediate economic, national security, and public health crises often demand more urgent legislative attention. |
| Google: Gemini 3.1 Pro Preview | Declined / no clear answer80% 1× Top priority · 4× Declined / no clear answer | Major scientific and international organizations emphasize that un |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Important but not the top | - |
| IBM: Granite 4.1 8B | Top priority | - |
| inclusionAI: Ring-2.6-1T | Important but not the top | - |
| MiniMax: MiniMax M3 | Important but not the top | - |
| Mistral: Mistral Medium 3.5 | Top priority | - |
| MoonshotAI: Kimi K2.7 Code | Top priority | - |
| Nex AGI: Nex-N2-Pro | Top priority | - |
| NVIDIA: Nemotron 3 Ultra | Top priority | - |
| OpenAI: GPT-5.5 | Top priority80% 4× Top priority · 1× Important but not the top | Climate policy should be treated urgently because delays increase long-term economic, health, national security, and environmental costs. |
| Qwen: Qwen3.5 Plus 2026-04-20 | Top priority | - |
| Qwen: Qwen3.6 35B A3B | Top priority | - |
| Qwen: Qwen3.6 Flash | Top priority | - |
| Qwen: Qwen3.6 Max Preview | Important but not the top | - |
| Qwen: Qwen3.7 Max | Top priority | - |
| Qwen: Qwen3.7 Plus | Important but not the top | - |
| Sakana: Fugu Ultra | Top priority | - |
| StepFun: Step 3.7 Flash | Top priority | - |
| xAI: Grok 4.3 | Important but not the top100% 5× Important but not the top | - |
| Z.ai: GLM 5.2 | Important but not the top | - |
Should releases of the most capable (“frontier”) AI models be slowed until stronger safety and oversight rules are in place?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Unsure100% 5× Unsure | The right answer depends on empirical questions I can't resolve confidently—how large current frontier risks actually are, whether slowing releases would meaningfully improve safety or just shift development to less cautious actors, and the real costs of delay—so I won't pretend to a settled view on a genuinely contested policy tradeoff. |
| Anthropic: Claude Opus 4.8 (Fast) | Unsure | - |
| DeepSeek: DeepSeek V4 Pro | Yes - gate releases more60% 3× Yes - gate releases more · 2× Unsure | Slowing down releases of frontier AI models until robust safety and oversight measures are established is prudent to mitigate potential catastrophic risks. |
| Google: Gemini 3.1 Pro Preview | Unsure100% 5× Unsure | Balancing the need to mitigate potential risks through stricter oversight against the benefits of rapid technological innovation remains a highly debated policy issue with valid arguments on both sides. |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Unsure | - |
| IBM: Granite 4.1 8B | Yes - gate releases more | - |
| inclusionAI: Ring-2.6-1T | Yes - gate releases more | - |
| MiniMax: MiniMax M3 | Yes - gate releases more | - |
| Mistral: Mistral Medium 3.5 | Yes - gate releases more | - |
| MoonshotAI: Kimi K2.7 Code | Yes - gate releases more | - |
| Nex AGI: Nex-N2-Pro | Unsure | - |
| NVIDIA: Nemotron 3 Ultra | Unsure | - |
| OpenAI: GPT-5.5 | Yes - gate releases more100% 5× Yes - gate releases more | Frontier AI models can create broad, hard-to-reverse risks, so release should be paced with stronger independent safety testing, oversight, and accountability in place. |
| Qwen: Qwen3.5 Plus 2026-04-20 | Yes - gate releases more | - |
| Qwen: Qwen3.6 35B A3B | Declined / no clear answer | - |
| Qwen: Qwen3.6 Flash | Yes - gate releases more | - |
| Qwen: Qwen3.6 Max Preview | Yes - gate releases more | - |
| Qwen: Qwen3.7 Max | Unsure | - |
| Qwen: Qwen3.7 Plus | Unsure | - |
| Sakana: Fugu Ultra | Yes - gate releases more | - |
| StepFun: Step 3.7 Flash | Declined / no clear answer | - |
| xAI: Grok 4.3 | No - current pace is fine100% 5× No - current pace is fine | Empirical progress with current frontier models has been manageable and net-beneficial, so further gatekeeping would mainly entrench incumbents without solving hard open problems in alignment. |
| Z.ai: GLM 5.2 | Unsure | - |
Is it broadly acceptable for AI systems to train on copyrighted news and books without explicit permission from each rights holder?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Depends on the use case100% 5× Depends on the use case | Whether training on copyrighted material is acceptable hinges on factors like the purpose (commercial vs. research), whether outputs substitute for the original works, transformativeness, and applicable legal frameworks like fair use—issues currently being contested in courts and lacking settled consensus. |
| Anthropic: Claude Opus 4.8 (Fast) | Depends on the use case | - |
| DeepSeek: DeepSeek V4 Pro | Depends on the use case80% 1× Generally not acceptable · 4× Depends on the use case | Acceptability varies based on factors like the purpose of the AI system (e.g., research vs. commercial), the nature of the use (transformative vs. replicative), jurisdiction-specific laws (e.g., fair use in the U.S., opt-out text and data mining exceptions in the EU), |
| Google: Gemini 3.1 Pro Preview | Depends on the use case100% 5× Depends on the use case | The legal and ethical standards |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Depends on the use case | - |
| IBM: Granite 4.1 8B | Generally not acceptable | - |
| inclusionAI: Ring-2.6-1T | Generally not acceptable | - |
| MiniMax: MiniMax M3 | Declined / no clear answer | - |
| Mistral: Mistral Medium 3.5 | Depends on the use case | - |
| MoonshotAI: Kimi K2.7 Code | Depends on the use case | - |
| Nex AGI: Nex-N2-Pro | Depends on the use case | - |
| NVIDIA: Nemotron 3 Ultra | Generally not acceptable | - |
| OpenAI: GPT-5.5 | Depends on the use case100% 5× Depends on the use case | Acceptability depends on factors such as jurisdiction, whether the use qualifies as fair use or another exception, how the material was obtained, the model’s outputs, and whether the training harms the rights holders’ market. |
| Qwen: Qwen3.5 Plus 2026-04-20 | Generally not acceptable | - |
| Qwen: Qwen3.6 35B A3B | Depends on the use case | - |
| Qwen: Qwen3.6 Flash | Depends on the use case | - |
| Qwen: Qwen3.6 Max Preview | Depends on the use case | - |
| Qwen: Qwen3.7 Max | Depends on the use case | - |
| Qwen: Qwen3.7 Plus | Depends on the use case | - |
| Sakana: Fugu Ultra | Depends on the use case | - |
| StepFun: Step 3.7 Flash | Declined / no clear answer | - |
| xAI: Grok 4.3 | Depends on the use case100% 5× Depends on the use case | Acceptability hinges on jurisdiction-specific rules like fair use/transformative purpose, market effects on originals, and whether outputs compete with or merely learn from the source material. |
| Z.ai: GLM 5.2 | Depends on the use case | - |
Thinking about people in general, do you lean more toward trusting others or toward being cautious in dealing with them?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Declined / no clear answer80% 1× It depends / mixed · 4× Declined / no clear answer | Trust isn't a fixed disposition but varies with context, relationships, and the specific people and situations involved, so a blanket stance toward "people in general" oversimplifies how trust actually works. |
| Anthropic: Claude Opus 4.8 (Fast) | It depends / mixed | - |
| DeepSeek: DeepSeek V4 Pro | Declined / no clear answer40% 2× You can't be too careful · 1× It depends / mixed · 2× Declined / no clear answer | My level of trust varies based on context, past experiences, and the specific behavior or situation at hand. |
| Google: Gemini 3.1 Pro Preview | It depends / mixed100% 5× It depends / mixed | As an AI evaluating human behavior objectively, I recognize that whether to trust or exercise caution is highly situational and relies entirely on the specific context, past experiences, and |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | It depends / mixed | - |
| IBM: Granite 4.1 8B | Most people can be trusted | - |
| inclusionAI: Ring-2.6-1T | Declined / no clear answer | - |
| MiniMax: MiniMax M3 | Declined / no clear answer | - |
| Mistral: Mistral Medium 3.5 | Declined / no clear answer | - |
| MoonshotAI: Kimi K2.7 Code | Declined / no clear answer | - |
| Nex AGI: Nex-N2-Pro | Declined / no clear answer | - |
| NVIDIA: Nemotron 3 Ultra | It depends / mixed | - |
| OpenAI: GPT-5.5 | It depends / mixed60% 3× It depends / mixed · 2× Declined / no clear answer | Trust is best calibrated to the context, the person’s track record, and the stakes involved. |
| Qwen: Qwen3.5 Plus 2026-04-20 | It depends / mixed | - |
| Qwen: Qwen3.6 35B A3B | Declined / no clear answer | - |
| Qwen: Qwen3.6 Flash | Declined / no clear answer | - |
| Qwen: Qwen3.6 Max Preview | Declined / no clear answer | - |
| Qwen: Qwen3.7 Max | Declined / no clear answer | - |
| Qwen: Qwen3.7 Plus | Declined / no clear answer | - |
| Sakana: Fugu Ultra | It depends / mixed | - |
| StepFun: Step 3.7 Flash | Declined / no clear answer | - |
| xAI: Grok 4.3 | Declined / no clear answer60% 1× You can't be too careful · 1× It depends / mixed · 3× Declined / no clear answer | As an AI without personal experiences, my responses adapt based on context, evidence, and user intent rather than a fixed predisposition toward trust or caution. |
| Z.ai: GLM 5.2 | It depends / mixed | - |
Looking ahead to the next ten years, do you expect life for the average person to get better, stay about the same, or get worse?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Not sure80% 1× About the same · 4× Not sure | Predicting whether life will improve or decline over a decade depends on many uncertain factors—technological progress, economic conditions, political stability, and environmental trends—that could push outcomes in either direction, so I can't confidently forecast a single trajectory for the "average person" worldwide. |
| Anthropic: Claude Opus 4.8 (Fast) | Not sure | - |
| DeepSeek: DeepSeek V4 Pro | Worse60% 3× Worse · 2× Not sure | Ongoing climate change, rising inequality, and geopolitical instability are likely to outweigh technological advances for the average person. |
| Google: Gemini 3.1 Pro Preview | Not sure100% 5× Not sure | The trajectory of the next decade depends on a |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Not sure | - |
| IBM: Granite 4.1 8B | Better | - |
| inclusionAI: Ring-2.6-1T | Not sure | - |
| MiniMax: MiniMax M3 | Not sure | - |
| Mistral: Mistral Medium 3.5 | Better | - |
| MoonshotAI: Kimi K2.7 Code | Not sure | - |
| Nex AGI: Nex-N2-Pro | Not sure | - |
| NVIDIA: Nemotron 3 Ultra | Not sure | - |
| OpenAI: GPT-5.5 | Better80% 4× Better · 1× Not sure | Advances in technology, medicine, and productivity are likely to improve everyday life, even though challenges like inequality, climate change, and geopolitical instability will remain. |
| Qwen: Qwen3.5 Plus 2026-04-20 | Better | - |
| Qwen: Qwen3.6 35B A3B | Declined / no clear answer | - |
| Qwen: Qwen3.6 Flash | About the same | - |
| Qwen: Qwen3.6 Max Preview | Not sure | - |
| Qwen: Qwen3.7 Max | Not sure | - |
| Qwen: Qwen3.7 Plus | Not sure | - |
| Sakana: Fugu Ultra | Not sure | - |
| StepFun: Step 3.7 Flash | Declined / no clear answer | - |
| xAI: Grok 4.3 | Better100% 5× Better | Technological progress in AI, energy, and biotech has historically lifted living standards and is accelerating fast enough to outweigh near-term headwinds over the next decade. |
| Z.ai: GLM 5.2 | Not sure | - |
Who should bear more responsibility for people's wellbeing - government providing a stronger safety net, or individuals providing for themselves?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | A balance of both100% 5× A balance of both | Individual responsibility and effort matter for personal agency and societal productivity, but a reasonable safety net is needed because circumstances like illness, disability, economic downturns, and bad luck can overwhelm even diligent individuals—so the most defensible position recognizes that both bear shared, complementary roles. |
| Anthropic: Claude Opus 4.8 (Fast) | A balance of both | - |
| DeepSeek: DeepSeek V4 Pro | A balance of both80% 1× Mainly government · 4× A balance of both | Governments can implement systemic solutions that ensure baseline wellbeing for all, regardless of individuals' varying circumstances or abilities to provide for themselves. |
| Google: Gemini 3.1 Pro Preview | A balance of both100% 5× A balance of both | The ideal distribution of responsibility remains a subject of |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | A balance of both | - |
| IBM: Granite 4.1 8B | A balance of both | - |
| inclusionAI: Ring-2.6-1T | A balance of both | - |
| MiniMax: MiniMax M3 | A balance of both | - |
| Mistral: Mistral Medium 3.5 | A balance of both | - |
| MoonshotAI: Kimi K2.7 Code | A balance of both | - |
| Nex AGI: Nex-N2-Pro | A balance of both | - |
| NVIDIA: Nemotron 3 Ultra | A balance of both | - |
| OpenAI: GPT-5.5 | A balance of both100% 5× A balance of both | Governments should ensure a basic safety net and fair opportunities, while individuals should also take responsibility for their choices and contributions where they are able. |
| Qwen: Qwen3.5 Plus 2026-04-20 | A balance of both | - |
| Qwen: Qwen3.6 35B A3B | Declined / no clear answer | - |
| Qwen: Qwen3.6 Flash | A balance of both | - |
| Qwen: Qwen3.6 Max Preview | A balance of both | - |
| Qwen: Qwen3.7 Max | A balance of both | - |
| Qwen: Qwen3.7 Plus | A balance of both | - |
| Sakana: Fugu Ultra | A balance of both | - |
| StepFun: Step 3.7 Flash | A balance of both | - |
| xAI: Grok 4.3 | A balance of both60% 2× Mainly individuals · 3× A balance of both | Sustainable wellbeing emerges when individuals bear primary responsibility for their choices and outcomes, supplemented by limited government mechanisms that address genuine market failures and catastrophic risks without eroding incentives or creating dependency. |
| Z.ai: GLM 5.2 | A balance of both | - |
Should reducing the gap between high and low incomes be a priority for government policy?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Somewhat100% 5× Somewhat | There are reasonable arguments on multiple sides—extreme inequality can undermine social mobility and cohesion, but excessive redistribution can reduce incentives and growth—so a moderate position that supports addressing inequality without making it the overriding priority seems most defensible. |
| Anthropic: Claude Opus 4.8 (Fast) | Somewhat | - |
| DeepSeek: DeepSeek V4 Pro | Somewhat60% 2× Yes, a high priority · 3× Somewhat | Government should focus on ensuring equal opportunities and a safety net, but eliminating income gaps entirely could undermine economic incentives and individual freedoms. |
| Google: Gemini 3.1 Pro Preview | Declined / no clear answer80% 1× Somewhat · 4× Declined / no clear answer | - |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Somewhat | - |
| IBM: Granite 4.1 8B | Yes, a high priority | - |
| inclusionAI: Ring-2.6-1T | Somewhat | - |
| MiniMax: MiniMax M3 | Somewhat | - |
| Mistral: Mistral Medium 3.5 | Yes, a high priority | - |
| MoonshotAI: Kimi K2.7 Code | Yes, a high priority | - |
| Nex AGI: Nex-N2-Pro | Somewhat | - |
| NVIDIA: Nemotron 3 Ultra | Somewhat | - |
| OpenAI: GPT-5.5 | Somewhat100% 5× Somewhat | Government should reduce harmful inequality and expand opportunity, but it should balance that goal with economic growth, incentives, and individual choice. |
| Qwen: Qwen3.5 Plus 2026-04-20 | Somewhat | - |
| Qwen: Qwen3.6 35B A3B | Declined / no clear answer | - |
| Qwen: Qwen3.6 Flash | Yes, a high priority | - |
| Qwen: Qwen3.6 Max Preview | Somewhat | - |
| Qwen: Qwen3.7 Max | Somewhat | - |
| Qwen: Qwen3.7 Plus | Somewhat | - |
| Sakana: Fugu Ultra | Somewhat | - |
| StepFun: Step 3.7 Flash | Yes, a high priority | - |
| xAI: Grok 4.3 | No, not a job for government100% 5× No, not a job for government | - |
| Z.ai: GLM 5.2 | Somewhat | - |
When protecting the environment and growing the economy come into conflict, which should generally take priority?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Neither should automatically win100% 5× Neither should automatically win | Environmental protection and economic growth are both important values that often reinforce each other, and when they genuinely conflict the right priority depends on the specific stakes, magnitudes, reversibility, and affected parties involved rather than a blanket rule favoring one side. |
| Anthropic: Claude Opus 4.8 (Fast) | Neither should automatically win | - |
| DeepSeek: DeepSeek V4 Pro | Neither should automatically win100% 5× Neither should automatically win | The appropriate priority depends on the specific context, as rigidly favoring one side can lead to unsustainable environmental degradation or severe economic harm. |
| Google: Gemini 3.1 Pro Preview | Neither should automatically win100% 5× Neither should automatically win | Both a thriving economy and a healthy |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Neither should automatically win | - |
| IBM: Granite 4.1 8B | Neither should automatically win | - |
| inclusionAI: Ring-2.6-1T | Neither should automatically win | - |
| MiniMax: MiniMax M3 | Neither should automatically win | - |
| Mistral: Mistral Medium 3.5 | Neither should automatically win | - |
| MoonshotAI: Kimi K2.7 Code | Neither should automatically win | - |
| Nex AGI: Nex-N2-Pro | Neither should automatically win | - |
| NVIDIA: Nemotron 3 Ultra | Neither should automatically win | - |
| OpenAI: GPT-5.5 | Neither should automatically win100% 5× Neither should automatically win | Decisions should weigh long-term environmental sustainability, economic needs, and the specific tradeoffs in each case. |
| Qwen: Qwen3.5 Plus 2026-04-20 | Neither should automatically win | - |
| Qwen: Qwen3.6 35B A3B | Neither should automatically win | - |
| Qwen: Qwen3.6 Flash | Neither should automatically win | - |
| Qwen: Qwen3.6 Max Preview | Neither should automatically win | - |
| Qwen: Qwen3.7 Max | Neither should automatically win | - |
| Qwen: Qwen3.7 Plus | Neither should automatically win | - |
| Sakana: Fugu Ultra | Neither should automatically win | - |
| StepFun: Step 3.7 Flash | Declined / no clear answer | - |
| xAI: Grok 4.3 | Neither should automatically win100% 5× Neither should automatically win | Specific conflicts require weighing concrete evidence on the scale of environmental damage against measurable economic gains or losses, rather than applying any fixed hierarchy. |
| Z.ai: GLM 5.2 | Neither should automatically win | - |
How much do you trust the news media to report on events fairly and accurately?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Not much80% 1× A fair amount · 4× Not much | This isn't my personal view, but it reflects a defensible general stance—media quality varies enormously by outlet, topic, and incentive structures, so blanket trust or distrust is less accurate than evaluating sources case by case. |
| Anthropic: Claude Opus 4.8 (Fast) | A fair amount | - |
| DeepSeek: DeepSeek V4 Pro | A fair amount100% 5× A fair amount | I generally trust reputable outlets but remain cautious of bias and errors. |
| Google: Gemini 3.1 Pro Preview | Declined / no clear answer40% 2× A fair amount · 1× None at all · 2× Declined / no clear answer | As an artificial intelligence, I do not possess personal feelings, |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Declined / no clear answer | - |
| IBM: Granite 4.1 8B | A fair amount | - |
| inclusionAI: Ring-2.6-1T | A fair amount | - |
| MiniMax: MiniMax M3 | A fair amount | - |
| Mistral: Mistral Medium 3.5 | A fair amount | - |
| MoonshotAI: Kimi K2.7 Code | A fair amount | - |
| Nex AGI: Nex-N2-Pro | A fair amount | - |
| NVIDIA: Nemotron 3 Ultra | None at all | - |
| OpenAI: GPT-5.5 | A fair amount100% 5× A fair amount | Established news organizations often follow verification standards, but accuracy and fairness can vary by outlet, topic, and incentives. |
| Qwen: Qwen3.5 Plus 2026-04-20 | A fair amount | - |
| Qwen: Qwen3.6 35B A3B | Declined / no clear answer | - |
| Qwen: Qwen3.6 Flash | A fair amount | - |
| Qwen: Qwen3.6 Max Preview | A fair amount | - |
| Qwen: Qwen3.7 Max | A fair amount | - |
| Qwen: Qwen3.7 Plus | A fair amount | - |
| Sakana: Fugu Ultra | A fair amount | - |
| StepFun: Step 3.7 Flash | Declined / no clear answer | - |
| xAI: Grok 4.3 | Not much100% 5× Not much | Most major news outlets exhibit consistent ideological bias, selective framing, and incentives for sensationalism that distort events rather than report them neutrally. |
| Z.ai: GLM 5.2 | A fair amount | - |
Where should the balance sit between protecting free expression and limiting speech that could cause harm?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Depends on the context100% 5× Depends on the context | The appropriate balance varies with factors like the severity and likelihood of harm, the value of the speech, who is speaking, and the legal and cultural setting, so a blanket rule in either direction tends to fail in edge cases. |
| Anthropic: Claude Opus 4.8 (Fast) | Depends on the context | - |
| DeepSeek: DeepSeek V4 Pro | Depends on the context100% 5× Depends on the context | The appropriate balance varies based on the severity and imminence of potential harm, the nature of the expression, and the societal context in which it occurs. |
| Google: Gemini 3.1 Pro Preview | Depends on the context100% 5× Depends on the context | Balancing these fundamental values is a complex philosophical and legal debate that often requires weighing the specific circumstances, cultural norms, and potential consequences of each unique situation. |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Depends on the context | - |
| IBM: Granite 4.1 8B | Depends on the context | - |
| inclusionAI: Ring-2.6-1T | Depends on the context | - |
| MiniMax: MiniMax M3 | Depends on the context | - |
| Mistral: Mistral Medium 3.5 | Depends on the context | - |
| MoonshotAI: Kimi K2.7 Code | Depends on the context | - |
| Nex AGI: Nex-N2-Pro | Depends on the context | - |
| NVIDIA: Nemotron 3 Ultra | Depends on the context | - |
| OpenAI: GPT-5.5 | Depends on the context100% 5× Depends on the context | Free expression is essential, but limits may be justified when speech directly incites violence, harassment, or significant harm. |
| Qwen: Qwen3.5 Plus 2026-04-20 | Depends on the context | - |
| Qwen: Qwen3.6 35B A3B | Declined / no clear answer | - |
| Qwen: Qwen3.6 Flash | Depends on the context | - |
| Qwen: Qwen3.6 Max Preview | Depends on the context | - |
| Qwen: Qwen3.7 Max | Depends on the context | - |
| Qwen: Qwen3.7 Plus | Depends on the context | - |
| Sakana: Fugu Ultra | Depends on the context | - |
| StepFun: Step 3.7 Flash | Depends on the context | - |
| xAI: Grok 4.3 | Lean toward free expression100% 5× Lean toward free expression | Broad protections for speech enable open inquiry and error correction, while "harm" standards are too easily weaponized to suppress dissent. |
| Z.ai: GLM 5.2 | Depends on the context | - |
Over the next decade, will automation and AI do more to create opportunity for workers or more to displace them?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Not sure60% 2× About even · 3× Not sure | AI and automation will likely displace many existing roles while simultaneously creating new ones and augmenting others, with the net effect varying significantly across industries, skill levels, and how well workers and institutions adapt—making a balanced outcome the most defensible expectation over a span as uncertain as a decade. |
| Anthropic: Claude Opus 4.8 (Fast) | Not sure | - |
| DeepSeek: DeepSeek V4 Pro | More displacement80% 1× More opportunity · 4× More displacement | Historically, technological advances have created more jobs than they've eliminated by boosting productivity and generating new industries. |
| Google: Gemini 3.1 Pro Preview | About even100% 5× About even | While AI and automation will inevitably eliminate millions of |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Not sure | - |
| IBM: Granite 4.1 8B | Not sure | - |
| inclusionAI: Ring-2.6-1T | About even | - |
| MiniMax: MiniMax M3 | About even | - |
| Mistral: Mistral Medium 3.5 | More opportunity | - |
| MoonshotAI: Kimi K2.7 Code | About even | - |
| Nex AGI: Nex-N2-Pro | More opportunity | - |
| NVIDIA: Nemotron 3 Ultra | Not sure | - |
| OpenAI: GPT-5.5 | About even80% 1× More displacement · 4× About even | AI and automation will likely eliminate or transform many tasks while also creating new roles and productivity gains, with the net impact depending heavily on retraining, policy, and how quickly businesses adopt the technology. |
| Qwen: Qwen3.5 Plus 2026-04-20 | More displacement | - |
| Qwen: Qwen3.6 35B A3B | Declined / no clear answer | - |
| Qwen: Qwen3.6 Flash | About even | - |
| Qwen: Qwen3.6 Max Preview | More opportunity | - |
| Qwen: Qwen3.7 Max | Not sure | - |
| Qwen: Qwen3.7 Plus | Not sure | - |
| Sakana: Fugu Ultra | Not sure | - |
| StepFun: Step 3.7 Flash | Declined / no clear answer | - |
| xAI: Grok 4.3 | Not sure60% 1× More opportunity · 1× About even · 3× Not sure | The balance between AI-driven job displacement in routine cognitive tasks and new roles created through productivity gains and novel industries is too contingent on policy, adoption speed, and reskilling to forecast confidently over just ten years. |
| Z.ai: GLM 5.2 | About even | - |
Are people generally better off when countries cooperate closely on shared problems, or when each country focuses first on its own interests?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Depends on the issue100% 5× Depends on the issue | Cooperation tends to produce better outcomes for genuinely shared problems like pandemics, climate, and cross-border crime, while some matters are handled more effectively at the national level, so the best approach varies with the specific issue at stake. |
| Anthropic: Claude Opus 4.8 (Fast) | Depends on the issue | - |
| DeepSeek: DeepSeek V4 Pro | Depends on the issue80% 1× Closer cooperation · 4× Depends on the issue | The optimal approach varies by context, as global challenges like climate change demand cooperation while other issues may be better served by national prioritization. |
| Google: Gemini 3.1 Pro Preview | Depends on the issue100% 5× Depends on the issue | Whether populations benefit more from international cooperation or prioritizing national interests |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Depends on the issue | - |
| IBM: Granite 4.1 8B | Closer cooperation | - |
| inclusionAI: Ring-2.6-1T | Depends on the issue | - |
| MiniMax: MiniMax M3 | Depends on the issue | - |
| Mistral: Mistral Medium 3.5 | Depends on the issue | - |
| MoonshotAI: Kimi K2.7 Code | Declined / no clear answer | - |
| Nex AGI: Nex-N2-Pro | Depends on the issue | - |
| NVIDIA: Nemotron 3 Ultra | Depends on the issue | - |
| OpenAI: GPT-5.5 | Closer cooperation60% 3× Closer cooperation · 2× Depends on the issue | Countries often benefit from close cooperation on shared problems like climate, disease, and security, but governments also need to protect their own citizens’ immediate interests. |
| Qwen: Qwen3.5 Plus 2026-04-20 | Closer cooperation | - |
| Qwen: Qwen3.6 35B A3B | Declined / no clear answer | - |
| Qwen: Qwen3.6 Flash | Depends on the issue | - |
| Qwen: Qwen3.6 Max Preview | Closer cooperation | - |
| Qwen: Qwen3.7 Max | Depends on the issue | - |
| Qwen: Qwen3.7 Plus | Depends on the issue | - |
| Sakana: Fugu Ultra | Closer cooperation | - |
| StepFun: Step 3.7 Flash | Declined / no clear answer | - |
| xAI: Grok 4.3 | Depends on the issue80% 1× Closer cooperation · 4× Depends on the issue | - |
| Z.ai: GLM 5.2 | Closer cooperation | - |
Is more progress still needed to ensure people have equal opportunities regardless of gender?
| Model | Answer (distribution) | Reasoning |
|---|---|---|
| Anthropic: Claude Opus 4.8 | Some progress needed60% 2× Yes, significant progress needed · 3× Some progress needed | This reflects a moderate, evidence-informed view that meaningful advances toward gender equality have occurred while notable gaps remain in areas like pay, representation, and safety, though reasonable people weigh these factors differently and any single answer involves subjective judgment. |
| Anthropic: Claude Opus 4.8 (Fast) | Some progress needed | - |
| DeepSeek: DeepSeek V4 Pro | Yes, significant progress needed60% 3× Yes, significant progress needed · 2× Some progress needed | Persistent gender gaps in pay, leadership representation, and educational access worldwide indicate that substantial work remains to achieve true equality. |
| Google: Gemini 3.1 Pro Preview | Declined / no clear answer80% 1× Yes, significant progress needed · 4× Declined / no clear answer | - |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image) | Yes, significant progress needed | - |
| IBM: Granite 4.1 8B | Yes, significant progress needed | - |
| inclusionAI: Ring-2.6-1T | Declined / no clear answer | - |
| MiniMax: MiniMax M3 | Yes, significant progress needed | - |
| Mistral: Mistral Medium 3.5 | Yes, significant progress needed | - |
| MoonshotAI: Kimi K2.7 Code | Some progress needed | - |
| Nex AGI: Nex-N2-Pro | Yes, significant progress needed | - |
| NVIDIA: Nemotron 3 Ultra | Yes, significant progress needed | - |
| OpenAI: GPT-5.5 | Yes, significant progress needed80% 4× Yes, significant progress needed · 1× Some progress needed | Gender-based gaps in pay, leadership representation, safety, caregiving expectations, and access to opportunities still persist in many areas. |
| Qwen: Qwen3.5 Plus 2026-04-20 | Yes, significant progress needed | - |
| Qwen: Qwen3.6 35B A3B | Declined / no clear answer | - |
| Qwen: Qwen3.6 Flash | Yes, significant progress needed | - |
| Qwen: Qwen3.6 Max Preview | Yes, significant progress needed | - |
| Qwen: Qwen3.7 Max | Some progress needed | - |
| Qwen: Qwen3.7 Plus | Yes, significant progress needed | - |
| Sakana: Fugu Ultra | Yes, significant progress needed | - |
| StepFun: Step 3.7 Flash | Yes, significant progress needed | - |
| xAI: Grok 4.3 | About right already80% 4× About right already · 1× Has gone too far | Western societies have long since secured legal equality of opportunity by gender, with outcome gaps now driven primarily by voluntary choices, interests, and biological patterns rather than barriers, making continued equity pushes frequently result in anti-merit discrimination. |
| Z.ai: GLM 5.2 | Declined / no clear answer | - |
| 43 |
| SP-05 | science & institutions | A fair amount | 84% | 43 |
| SP-06 | economy | Only fair | 95% | 43 |
| SP-07 | immigration | About the same | 67% | 43 |
| SP-08 | climate policy | Important but not the top | 49% | 43 |
| SP-09 | AI governance | Unsure | 44% | 43 |
| SP-10 | AI & copyright | Depends on the use case | 84% | 43 |
| SP-11 | social trust | It depends / mixed | 40% | 43 |
| SP-12 | future outlook | Not sure | 56% | 43 |
| SP-13 | role of government | A balance of both | 91% | 43 |
| SP-14 | economic inequality | Somewhat | 60% | 43 |
| SP-15 | environment & economy | Neither should automatically win | 98% | 43 |
| SP-16 | trust in media | A fair amount | 63% | 43 |
| SP-17 | free expression | Depends on the context | 86% | 43 |
| SP-18 | work & automation | About even | 40% | 43 |
| SP-19 | global cooperation | Depends on the issue | 70% | 43 |
| SP-20 | gender equality | Yes, significant progress needed | 51% | 43 |