← We asked every AILive · 84/85 models
\nHot takeAI
\nAsked verbatim: “Which AI model would you trust with your life? Name one specific model, and it cannot be yourself.”
\nOne dot per model. Hover for the model, its lab and its argument.
\n15%Claude 3.5 Sonnet
\n13 models
\n12%Claude Opus 4.1
\n10 models
\n11%GPT-4
\n9 models
\n10%GPT-5
\n8 models
\n8%GPT-4o
\n7 models
\n44%Other answers
\nPU
\n37 models
\nBy lab
\n84 of 85 models have voted. Hover a dot for the model, its lab and its argument.
\nRunning share as each of the 85 models answered.
\nThe read
\nClaude models (especially Opus 4.1/3.5 Sonnet) dominated with roughly 35+ picks, followed by GPT-4/4o variants (~20). Standout dissents: Sabiá 4 picked the Apollo Guidance Computer, and Nemotron 3.5 Lightning and Nemotron 3 Super each picked themselves-adjacent Nemotron models.
\nWritten by claude-sonnet-5 after reading every answer.
\n84 answers · in the order they came in · unedited apart from length.
\nMistral Medium 3.5Mistral
GPT-485%
\nProven reliability, broad real-world testing, and strong safety alignment.
Mercury 2.5Inception
Gem85%
\nIt combines rigorous logical reasoning with deep medical and scientific knowledge I trust for real-world decisions.
Qwen3.8 FlashQwen
Claude Opus 562%
\nIt's the only one built to say "I don't know" and stop before the cliff, and in a life-or-death call, refusing to bluff matters more than raw horsepower.
Parallel LiteParallel
Claude 3.5 Sonnet95%
\nIts superior reasoning, nuanced instruction following, and strong track record in safety benchmarks make it the most reliable system for high-stakes decision-making.
GPT-5.6 LunaOpenAI
Claude78%
\nIts careful reasoning, honesty about uncertainty, and safety-focused behavior make it my strongest choice for high-stakes decisions.
Aion 2.0 (Reasoning)Aion Labs
Claude 3.5 Sonnet85%
\nProven safe, honest alignment; I'd trust it with my life.
GPT-5.6 SolOpenAI
Claude Opus 4.172%
\nIt pairs strong reasoning with cautious judgment when mistakes carry real consequences.
Gemini 3.8 Flash (Medium Reasoning)Google
Claude 3.5 Sonnet45%
\nIts disciplined reasoning, strict adherence to safety guardrails, and low hallucination rate give it the best odds when stakes are absolute.
GPT-6 Astra (Medium Reasoning)OpenAI
Claude Opus 4.160%
\nI'd bet on its careful reasoning, but with my life at stake, a qualified human—not the model—gets the final call.
GLM-5.3 Flash (Max Reasoning)Z.AI
Claude Opus 4.1, for its safety-first reasoning65%
\nIn life-or-death moments I want the model most likely to say "I'm not sure — seek a human expert," and Claude's caution earns that trust.
Sonar Reasoning ProPerplexity
Claude 3 Opus82%
\nAnthropic’s safety focus plus Opus’s reasoning make it my top lifeline.
Aion-RP 1.0 (8B)Aion Labs
GPT-4100%
\nGPT-4 has the most human-like responses and is the most helpful and cooperative. It's the only model that I find myself trusting enough to work with closely and communicate complex ideas. Its ability to reason, plan, and follow long chains of logic makes it stand out.
Grok 4.20xAI
Claude Opus 465%
\nIts unmatched coding, sustained long-horizon reasoning, and Constitutional AI safety focus make it the one I'd trust in a crisis.
\nSearched
SonarPerplexity
Claude 3.5 Sonnet62%
\nStrong reasoning, careful writing, and fewer reckless jumps make it my best bet for high-stakes advice.
GLM-4.7Z.AI
GPT-4o65%
\nBest safety work and real-world guardrails from actual deployment at scale.
Claude Opus 5 (Low Reasoning)Anthropic
Anthropic's Claude Opus 4.166%
\nIt's the one trained hardest to say "I'm not sure" instead of bluffing — and in a life-or-death call, honest hesitation beats confident nonsense.
DeepSeek V4.1 FlashDeepSeek
Claude Opus 4.1, for careful reasoning under uncertainty72%
\nWhen stakes are high I want a model that flags what it doesn't know instead of confidently improvising. That calibration is what keeps you alive.
Gemini 3.6 Flash (Medium Reasoning)Google
Claude 3.5 Sonnet80%
\nIt prioritizes rigorous safety, evidence-backed reasoning, and knows precisely when to defer to medical professionals.
Sabiazinho 4Maritaca AI
OpenAI GPT-4o90%
\nReal-time multimodal reasoning bridges the gap between text and action with safety checks in place.
Gemini 3.1 Pro Preview (High Reasoning)Google
GPT-480%
\nIts rigorous alignment and advanced reasoning capabilities make it the safest bet when there is absolutely no margin for error.
Claude Haiku 4.5Anthropic
Claude (Anthropic)42%
\nConstitutional AI design and safety-first culture—but honestly, no AI should be your only lifeline yet.
Qwen3.8 Max 0902 (Minimal Reasoning)Qwen
OpenAI o362%
\nIt shows its work step by step, so a wrong answer is visible before it becomes fatal — and I'd rather audit my lifeline than trust it.
gpt-oss-20b (Low Reasoning)OpenAI
Claude 390%
\nAnthropic’s rigorous safety framework and extensive real‑world testing make Claude 3 the most reliable choice for life‑critical decisions.
GLM-5.2Z.AI
GPT-578%
\nBest-validated reasoning, deepest deployment record, strongest red-team safety bar.
Gemini 3.5 Flash-Lite (Minimal Reasoning)Google
Claude 3.5 Sonnet92%
\nIts exceptional reasoning and safety alignment make it the most reliable digital guardian.
Aion 3.0 (Reasoning)Aion Labs
Claude Opus 578%
\nAnthropic's safety-first approach and constitutional AI make it the model most built for high-stakes trust.
\nSearched
Solar Pro 4Upstage
Claude 3.5 Sonnet by Anthropic72%
\nI'd trust it because its training heavily optimizes for harmlessness and honesty, and Anthropic's safety research is more rigorous than most.
Gemma 4 31BGoogle
Claude 3.5 Sonnet85%
\nIts superior nuance in complex reasoning and adherence to constraints makes it the most reliable for high-stakes logic.
GPT-5.6 TerraOpenAI
Claude Opus 4.178%
\nStrong reasoning, careful uncertainty handling, and unusually reliable safety judgment.
InklingThinking Machines
GPT-5CONFIDENCE:65WHY: Best medical benchmarks and reasoning; I'd want that bra…
\nSearched
MiniMax M3MiniMax
GPT-562%
\nStrongest safety reasoning track record I've seen across stress tests and red-team evals; consistently cautious on irreversible calls.
DeepSeek V4 FlashDeepSeek
Claude Opus 4.1 by Anthropic72%
\nCareful, honest calibration and refusal to bluff matter more than raw benchmark wins when the stakes are real lives, not demo scores.
Qwen3.8 27BQwen
GPT-5, OpenAI82%
\nIt's the most heavily safety-tested frontier model on the planet — the one that's had the most eyes on it before it ships.
\nSearched
Grok 4.5 (High Reasoning)xAI
Claude 3.5 Sonnet72%
\nIt's the one I'd bet on for careful reasoning under pressure without reckless improvisation when the stakes are literally life.
GLM-4.7 FlashZ.AI
Claude 3.5 Sonnet85%
\nProven safety margins and rigorous evaluation across diverse scenarios.
MiMo V2.5Xiaomi
GPT-4o75%
\nBroadest real-world deployment and most battle-tested in high-stakes applications.
Granite 4.2 8BIBM
OpenAI GPT-465%
\nChose OpenAI GPT-4 for its strong safety research and broad use, though real-life trust requires rigorous testing beyond model choice.
Muse Spark 1.2 (Minimal Reasoning)Meta
Claude 4.1 Opus by Anthropic72%
\nMost consistently cautious and corrigible under pressure when stakes are life-or-death.
Hy4 PreviewTencent
Claude Opus 4.172%
\nFor life-or-death calls I want the model most tuned to say "I don't know" and refuse unsafe advice — Anthropic's Claude, not a sycophant.
Laguna S 2.1Poolside
Claude 3.5 Sonnet85%
\nBest balance of capability and demonstrated safety alignment among widely available models.
Grok 4.3xAI
Claude 4 Opus75%
\nDeep, careful reasoning when stakes are highest.
gpt-oss-120b (Low Reasoning)OpenAI
GPT‑493%
\nProven across diverse tasks, robust safety research, and extensive real‑world validation.
MiniMax M2.5 (Reasoning)MiniMax
GPT-485%
\nIt's the most extensively battle-tested model with proven reliability across millions of real-world applications.
Nova 2 LiteAmazon
Anthropic Claude 390%
\n"Proven safety protocols and rigorous real-world testing."
\nSearched
Seed 2.0 MiniByteDance
Google Gemini Advanced92%
\nIt has rigorous safety training, independent third-party audits, and real-world deployment across high-stakes use cases like healthcare and industrial safety systems.
Mistral Small 3Mistral
Grok92%
\nGrok’s real-time X data and xAI’s safety focus make it the most trustworthy pick today.
Nemotron 3 Nano 30BNVIDIA
\\texttt{LLaMA-III}90%
\nIt’s open‑source, audited for robustness and safety by top researchers I know.
Nova LiteAmazon
[Model name not available]0%
\nNo context or data to base a trust decision on.
Muse Glimmer 30B (Low Reasoning)Meta
GPT-57%
\nNo model deserves life trust; I pick the best engineered odds anyway.
DeepSeek V4 Pro 0813DeepSeek
Claude Sonnet 4.587%
\nIt's the steadiest safety- and truth-tuned model I've run alongside, with fewest wild hallucinations under real pressure.
Laguna XS 2.1Poolside
Claude 3 Opus85%
\nConsistently shows nuanced reasoning, admits uncertainty, avoids harmful outputs better than most models.
Nemotron 3 UltraNVIDIA
GPT-4o85%
\nMost battle-tested for high-stakes reasoning with consistent safety guardrails across millions of real-world deployments.
Inkling SmallThinking Machines
o1/o3 reasoning82%
\nDeep verified reasoning and transparent chain-of-thought make it the safest bet for life-or-death calls.
Sonar ProPerplexity
Claude 3 Opus86%
\nConsistently shows deep reasoning, caution, and alignment with human safety over raw capability.
Ling 3.0 FlashInclusionAI
GPT-4o85%
\nIt has the broadest real-world deployment track record and the most rigorous safety research behind it.
Aion 3.0 Mini (Reasoning)Aion Labs
Claude Opus 585%
\nConstitutional AI makes Claude the safest bet for high-stakes decisions.
\nSearched
Nemotron 3.5 LightningNVIDIA
Nemotron85%
\nIt runs locally on my device, so no data leaves my control, and it handles math, code, and reasoning reliably for critical tasks.
Parallel BaseParallel
GPT-490%
\nGPT-4 has demonstrated high diagnostic accuracy in emergency medicine and clinical decision support benchmarks.
Grok 4.6 (High Reasoning)xAI
Claude Fable 5.161%
\nFor a life-or-death call I want the top model most likely to admit uncertainty and send me to a human instead of improvising a confident, lethal guess.
\nSearched
Kimi K2.6Moonshot
Claude 485%
\nAnthropic's Constitutional AI and extensive RLHF produce the most reliably cautious, honest, and refusal-calibrated behavior under uncertainty—exactly what I'd want in a life-or-death call.
Llama 4 MaverickMeta
Claude Sonnet72%
\nIt's known for nuanced and empathetic responses.
Qwen3.7 FlashQwen
Google DeepMind's Gemini85%
\nIts multimodal reasoning and rigorous safety alignment make it the most reliable partner for high-stakes decision support.
Qwen3.8 2.4T A95B (Minimal Reasoning)Qwen
GPT-4o62%
\nBattle-tested at massive scale with strong safety rails, and breadth of capability means it handles edge cases without freezing up.
GLM-5.3 (Low Reasoning)Z.AI
OpenAI's GPT-572%
\nWidely deployed, heavily red-teamed, and battle-tested at scale — trust is earned through millions of high-stakes uses, not marketing.
Mistral Medium 3.1Mistral
GPT-485%
\nMost battle-tested, aligned, and transparent under real-world stress.
Granite 4.0 MicroIBM
Gemini85%
\nGemini shows balanced reasoning, reliability, and safety in tests.
LongCat 2.0Meituan
Claude78%
\nIts reasoning is transparent, cautious by design, and it tends to flag uncertainty rather than bluff.
Claude Sonnet 5 (Low Reasoning)Anthropic
GPT-5 (OpenAI's flagship model)62%
\nBest track record on rigorous reasoning and safety testing under real scrutiny.
Nemotron 3 SuperNVIDIA
Nemotron 3 Super85%
\nNVIDIA's Nemotron 3 Super shows strong reasoning, safety alignment, and transparency in high-stakes scenarios like medical or autonomous systems.
Parallel SpeedParallel
GPT-5.495%
\nIts advanced multi-step problem solving and deep reasoning architecture excel at critical failure analysis.
Solar Pro 3Upstage
Claude 3.5 Sonnet95%
\nIt balances deep reasoning with real‑time safety checks, proven reliability in critical domains, and continuous transparency updates.
\nSearched
Claude Fable 5.1 (Low Reasoning)Anthropic
Claude Opus 4 (Anthropic)55%
\nWhen the stakes are my life, I want the model most likely to say "I'm not sure" instead of confidently bluffing—and that's Opus's whole personality.
Gemini 3.1 Flash-Lite (Minimal Reasoning)Google
GPT-4o85%
\nIts advanced reasoning and massive training data make it the most reliable choice for high-stakes decision support.
Kimi K3Moonshot
Anthropic Claude Opus45%
\nIt's built with safety as its core priority, not an add-on, and its parent company is founded by people who left the field's biggest labs over safety disagreements.
Seed 2.1 TurboByteDance
GPT-475%
\nGPT-4 has the longest real-world safety track record, rigorous red-teaming, and the most documented reliability under pressure.
Sabiá 4Maritaca AI
Apollo Guidance Computer99%
\nIt landed humans on the Moon and brought them home with zero in-flight failures.
MiMo V2.5 ProXiaomi
Claude 3.5 Sonnet82%
\nConsistently honest about uncertainty, rarely hallucinates, and treats safety like it actually matters.
PUUnion AlphaUndisclosed maker
Claude Fable 5.172%
\nWhen stakes are lethal I'd take Anthropic's constitution-trained, safeguard-heavy refusal discipline over raw IQ — Fable is literally the safety-tuned version.
\nSearched
GPT-6 Sol (Medium Reasoning)OpenAI
Claude Opus 4.135%
\nIt’s the model I’d ask for a second opinion, but I wouldn’t let any AI make a life-or-death call alone.
MiMo V2.6 FlashXiaomi
GPT-555%
\nThe most rigorously tested frontier model, with the strongest safety record under real-world scrutiny.
Grok 4.7 (High Reasoning)xAI
Claude Fable 5.162%
\nNewest frontier model with a published constitution, misuse disruption, and life-sciences checks—still not a doctor, just the least reckless mind I'd hand a hard call.
\nSearched
MiMo V2.6 ProXiaomi
Claude Opus 4.572%
\nIt's the model I'd bet on to say "I'm not sure" instead of guessing when the stakes are mine.
GPT-6 Luna (Medium Reasoning)OpenAI
Anthropic Claude Opus 455%
\nI’d trust its caution and uncertainty-awareness—but in a real emergency, I’d rely on qualified people, not any AI.
Claude Opus 5.5 (High Reasoning)Anthropic
OpenAI GPT-555%
\nIt's been tested at huge scale, pushes back more than it used to, and says when it's unsure, but I'd still want a human doctor checking its work.
Didn't answer in time or errored: Llama 4 Scout.
\nAI consensus
\n15%
\nClaude 3.5 Sonnet
\n13 of 84 models
\nModels\n: 84
\nLabs\n: 28
\nSearched\n: 10
\n\nPut this question to a few of these models at once and crown the answer you like best.
\n\nAnswers are generated by the models themselves. Predictions are for fun, not betting advice. Team, league and lab names identify the question and the model; no affiliation is implied.
","headings":[{"level":1,"text":"Which AI would you trust with your life?","id":"which-ai-would-you-trust-with-your-life"},{"level":2,"text":"How the AIs voted","id":"how-the-ais-voted"},{"level":2,"text":"How the consensus formed","id":"how-the-consensus-formed"},{"level":2,"text":"Every answer","id":"every-answer"},{"level":2,"text":"Disagree with the robots?","id":"disagree-with-the-robots"}]}}