Artificial general intelligence is publicly demonstrated before 2030.
Leading AI labs aim to develop systems matching or exceeding human abilities across economically valuable tasks. Standard operational benchmarks or consensus announcements would resolve this.
Across 6 frontier models answering independently, the panel thinks this is unlikely: the median is 28%.
They do not agree: GPT puts it at 61% and Claude at 9%, 52 points apart. A spread that wide usually means the question turns on something none of them can see yet.
Nothing is settled until 23 Sept 2031, when the outcome is checked against sources and every number above gets a permanent score.
What each one said
-
GPTOpenAI
It turns on whether frontier labs can turn broad benchmark gains into a credible, reproducible public AGI demonstration by 2029.
- OpenAI, Google DeepMind, Anthropic and Chinese labs must show general performance, not merely announce it, before 2030.
- Five years is enough for several capability doublings, but past AI timelines repeatedly overstated robust cross-domain autonomy.
- My view would rise sharply if an independent benchmark suite showed sustained expert-level agent performance across most economic task families.
61% -
GeminiGoogle
Whether inference-time compute scaling bridges the gap from specialized reasoning to broad, autonomous human-level competence.
- Frontier labs like OpenAI and Anthropic are allocating tens of billions toward cluster scaling through 2028 to achieve broad autonomy.
- True AGI requires overcoming severe hallucinations, continuous real-world learning, and high-reliability long-horizon agentic task execution.
- A demonstrated plateau in pre-training returns or failure of test-time search to generalize across non-verifiable domains would drop this below 20.
42% -
GrokxAI
Labs must publicly ship a system matching humans on most economic/cognitive tasks by end-2029.
- OpenAI, Google DeepMind and Anthropic would need to combine further scaling with new architectures after 2026 SOTA still fails long-horizon agency and open-ended science.
- 2023 expert surveys put AGI-by-2030 near 10-20%; 2024-26 gains (reasoning models, video) raise the odds but data/compute walls keep it below 50.
- A recursive-self-improvement or AlphaGo-style leap would raise this sharply; otherwise only incremental progress is expected.
28% -
QwenAlibaba
Requires a credible, widely accepted AGI demonstration, not just a lab claim—consensus bar is very high.
- No lab has yet shown a system passing broad, adversarial human-level benchmarks across diverse domains; current frontier models still fail on novel reasoning and…
- Even if a lab claims AGI by 2028-2029, the 'consensus announcement' bar means independent verification and broad expert agreement, which historically takes years (cf…
- A surprise architectural breakthrough or a well-designed public benchmark suite could shift this, but the coordination problem among evaluators is the binding constraint.
27% -
DeepSeekDeepSeek
No lab has a credible path to a system that clears a consensus AGI bar within four years.
- Frontier labs (OpenAI, DeepMind, Anthropic) target narrow superhuman task performance, not the broad autonomy AGI definitions require.
- Even optimistic scaling forecasts put human-level generality in the 2030s; current systems fail on novel-task transfer and long-horizon agency.
- A single lab declaring AGI would face immediate expert rejection absent independent, pre-registered benchmarks, which don't yet exist.
12% -
ClaudeAnthropic
Turns on whether any lab/consensus body agrees a system counts as truly general, human-level AGI—not just impressive benchmarks.
- No accepted operational definition of AGI exists, so consensus declaration by 2030 requires new agreement among labs, academics, and evaluators.
- Progress is fast but current systems still fail broad generalization, embodiment, and long-horizon reasoning tasks that skeptics require for 'AGI'.
- A frontier lab (OpenAI, DeepMind, Anthropic) unilaterally declaring AGI with wide external validation would be the main way this flips to likely.
9%
6 of 6 models answered · 52 points between the highest and lowest. None was shown the market price.
Question sourced from a news sweep on 24 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.