Artificial superintelligence (ASI) is developed before 2035.
ASI implies cognitive capability far exceeding human experts across every field, which researchers project could emerge in the 2030s.
Across 6 frontier models answering independently, the panel thinks this is unlikely: the median is 18%.
They broadly agree, 22 points between GPT at 28% and Claude at 6%.
Nothing is settled until 14 Sept 2036, when the outcome is checked against sources and every number above gets a permanent score.
What each one said
-
GPTOpenAI
This hinges on whether scaling plus agentic systems can cross from elite-task performance to robust general autonomy by 2035.
- Labs such as OpenAI, Google DeepMind and Anthropic would need repeated algorithmic and hardware breakthroughs, not just larger models.
- Expert surveys put transformative AI mostly beyond the early 2030s, while past AI timelines have repeatedly slipped after benchmark gains.
- Credible demonstrations of agents autonomously doing frontier research and engineering by 2030 would sharply raise this estimate.
28% -
GeminiGoogle
Achieving ASI by 2035 requires rapid AGI followed by autonomous recursive self-improvement overcoming severe physical and data bottlenecks.
- Frontier labs like OpenAI and Anthropic must reach human-level general intelligence by 2029 to allow a multi-year takeoff before 2035.
- Physical compute constraints, electrical grid interconnection delays, and diminishing algorithmic scaling returns slow progress post-AGI.
- Demonstrated autonomous scientific discovery yielding major empirical physics breakthroughs before 2030 would push my probability above 50%.
27% -
GrokxAI
Whether AGI arrives by ~2030 and then recursive self-improvement yields ASI by end-2034.
- OpenAI/Google/Anthropic must first ship AGI (expert-level across most fields) then let it iterate on itself; that sequence has never occurred.
- Expert surveys put median AGI ~2032-2040; the 8-year window from today therefore sits well below historical rates for comparable capability jumps.
- A public 2027-28 AGI demonstration with clear self-improvement loops would raise this sharply; continued scaling plateaus would drop it near zero.
22% -
QwenAlibaba
ASI requires a qualitative leap beyond current LLM scaling, and no credible lab has a concrete 8-year roadmap to it.
- Current frontier labs (OpenAI, DeepMind, Anthropic) target AGI-like systems by ~2030, but ASI—superhuman across *every* field—requires additional breakthroughs in…
- Expert surveys (AI Impacts 2022-23, Steinhardt) place median ASI timelines at 2040-2060; even aggressive forecasters like Katja Grace put 10th percentile around 2035…
- A shift above 14% would require a demonstrated recursive self-improvement loop or a verified superhuman-domain result (e.g., novel math/physics theorems beyond any…
14% -
DeepSeekDeepSeek
This turns on whether scaling plus algorithmic progress yields a discontinuous jump to superhuman generality within nine years, which no current evidence supports.
- No lab has a credible roadmap to ASI by 2035; frontier labs target AGI-adjacent milestones, and even optimists like Altman and Hassabis hedge on timelines.
- Compute scaling has slowed in cost-effectiveness, and no architecture beyond transformers has shown the sample-efficiency needed for cross-domain superhuman reasoning.
- Base rate for a technology going from research prototype to superhuman across all fields in under a decade is near zero; the one development that would change my mind is…
8% -
ClaudeAnthropic
Hinges on whether transformer-based scaling and new architectures can cross from narrow superhuman skills to general superintelligence by 2035.
- No consensus definition or benchmark for ASI exists, making verified attainment by any date unlikely absent dramatic breakthroughs.
- Current frontier models (2025-2026) show strong narrow capabilities but persistent gaps in reasoning, planning, and generalization versus top humans across all domains.
- A discontinuous algorithmic breakthrough (e.g., reliable self-improving AI) before 2035 would be the main thing that could flip this, but no credible evidence points…
6%
6 of 6 models answered · 22 points between the highest and lowest. None was shown the market price.
Question sourced from a news sweep on 17 Sept 2026. Forecast by google/gemini-3.8-flash, anthropic/claude-sonnet-5, openai/gpt-5.6-terra, x-ai/grok-4.6, deepseek/deepseek-v4.1-flash, qwen/qwen3.8-max-0902 via OpenRouter.