Artificial General Intelligence (AGI) is publicly demonstrated before 2030.
Leading frontier AI labs claim AGI is possible within years. This resolves if an AI matches human-level proficiency across virtually all economically and intellectually valuable tasks by 2030.
Across 6 frontier models answering independently, the panel thinks this is unlikely: the median is 15%.
They do not agree: Gemini puts it at 37% and Claude at 7%, 30 points apart. A spread that wide usually means the question turns on something none of them can see yet.
Nothing is settled until 9 Oct 2031, when the outcome is checked against sources and every number above gets a permanent score.
What each one said
-
GeminiGoogle
Solving long-horizon planning and real-world physical or novel scientific reasoning remains a steep hurdle within 3.5 years.
- Frontier labs like Anthropic and OpenAI face data saturation and power constraints, making purely software-driven AGI by 2029 hard.
- Matching 'virtually all' tasks requires flawless autonomy in open-ended work, whereas current systems still hallucinate and fail.
- A demonstrated breakthrough in self-improving recursive reasoning architectures by late 2027 would push these odds above 60.
37% -
GPTOpenAI
It hinges on whether scaling plus agents produces robust cross-domain autonomy, not impressive benchmark performance.
- Labs such as OpenAI, Google DeepMind and Anthropic can deploy demos quickly, but the bar requires breadth across nearly all valuable work.
- Past AI forecasts routinely overstate near-term generality; systems still fail at long-horizon reliability, tacit knowledge and real-world execution.
- A convincing agent that independently performs months of varied professional work by 2028 would sharply raise the odds.
28% -
GrokxAI
Whether 3 years of scaling plus algorithms close remaining gaps in agency, reliability and novel work.
- Frontier labs must publicly show a system matching humans on virtually all economic/intellectual tasks by end-2029, not just chat or coding.
- Expert surveys (HLMI medians 2040s) and 2023-26 progress (impressive but brittle) put this well below base rate.
- A 2027 system independently producing novel PhD research or running a firm would raise the odds sharply.
18% -
DeepSeekDeepSeek
No lab has a credible path to a single system matching humans across virtually all valuable tasks within four years.
- Frontier labs' own definitions (OpenAI, DeepMind) require broad autonomy and generalization that current LLMs lack; no benchmark or demo has shown cross-domain parity.
- Even optimistic timelines from Altman, Hassabis, and Amodei describe AGI as a 2027-2035 range, and their public demos consistently underdeliver on the 'virtually all…
- The base rate for grand AI milestones arriving on a specific 4-year schedule is near zero; the 2023-2026 period produced impressive narrow systems, not general ones.
12% -
QwenAlibaba
AGI requires matching humans across virtually all economically valuable tasks, not just benchmarks—a bar no lab is close to clearing.
- Frontier labs like OpenAI and Anthropic define AGI loosely in public but the resolution criterion here is 'virtually all' tasks, which is far stricter than current…
- Even optimistic internal timelines (Altman's 'we know how to build AGI') target narrow definitions; no system today handles physical reasoning, novel scientific…
- A surprise architectural breakthrough or a dramatic scaling result in 2027-2028 would be the main path to raising this, but the base rate of paradigm shifts in a 4-year…
12% -
ClaudeAnthropic
Hinges on whether any lab achieves broad human-level competence across nearly all cognitive/economic tasks, not just benchmarks.
- No consensus-accepted AGI demonstration exists today despite GPT-4/5-class models, and the gap to 'virtually all' tasks remains large in robotics, long-horizon planning…
- Labs like OpenAI and DeepMind have repeatedly pushed timelines and hyped progress for funding/PR reasons, which historically overstates actual arrival dates.
- A clear falsification standard (matching humans across almost all valuable tasks) is very high; narrow superhuman performance in coding or math won't satisfy it, keeping…
7%
6 of 6 models answered · 30 points between the highest and lowest. None was shown the market price.
Question sourced from a news sweep on 10 Oct 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.