An artificial intelligence system passes a consensus benchmark of human-level capabilities across major domains before 2030.
Public curiosity often focuses on when advanced AI will arrive. This tracks whether leading labs demonstrate systems matching median adult human capabilities across diverse cognitive domains by 2030.
Across 6 frontier models answering independently, the panel thinks this is unlikely: the median is 33%.
They do not agree: GPT puts it at 58% and DeepSeek at 12%, 46 points apart. A spread that wide usually means the question turns on something none of them can see yet.
Nothing is settled until 12 Nov 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 labs can convert fast model gains into robust, independently verified cross-domain competence.
- Frontier labs already field strong coding, science, language and tool-use systems, and four years permits several major training cycles.
- The bottleneck is a credible consensus benchmark covering generalization, long-horizon work and real-world reliability, not isolated test scores.
- This rises sharply if a broad third-party suite shows median-adult performance with reproducible results by 2028; safety or scaling stalls would cut it.
58% -
GeminiGoogle
Success hinges on establishing an agreed cross-domain benchmark and solving persistent reasoning and sample-efficiency bottlenecks.
- No authoritative body like NIST or ISO has yet established a binding consensus battery covering all median human adult capabilities.
- Frontier labs are saturated on narrow academic tests, but multi-hour planning, physical world models, and novel learning lag behind.
- A formal standard accepted by OpenAI, Anthropic, and independent academics by 2028 would sharply increase the likelihood of passage.
42% -
QwenAlibaba
The bottleneck is whether the field coalesces around a single benchmark, not whether the capability exists.
- No consensus benchmark currently exists; METR, ARC-AGI, and similar efforts are fragmented, and the field historically moves goalposts when AI passes tests.
- Technical capability is on track—frontier models already exceed median human on most knowledge-work benchmarks—but 'across major domains' including physical reasoning…
- A single development that would raise this sharply: a major lab publishing a comprehensive eval suite (like a 'Turing Test 2.0') that gets >70% citation adoption within…
38% -
GrokxAI
Labs must ship and a community must ratify a multi-domain human-median eval that a model then clears by end-2029.
- OpenAI/Google/Anthropic need robust long-horizon agency plus multimodal competence by ~2028; no consensus suite exists today.
- Capability jumps (GPT-3 to o1) imply 20-30% chance of closing the remaining gap in four years, but compute/data walls and lagging evals keep odds below 50.
- A public multi-lab eval (HELM-AGI style) adopted in 2027 would most raise this forecast.
28% -
ClaudeAnthropic
Hinges on whether any benchmark suite gets broad expert/community consensus as a true cross-domain human-parity test before 2030.
- No current benchmark (MMLU, GPQA, ARC-AGI-2, etc.) has consensus status as 'human-level across major domains'; definitions remain contested.
- AI progress is fast but breadth/generalization, long-horizon planning, and embodied reasoning still lag narrow benchmark performance, and goalposts keep moving.
- A major lab (OpenAI, DeepMind, Anthropic) or academic body coalescing around one authoritative benchmark with clear passing would be the key change signal.
22% -
DeepSeekDeepSeek
No agreed 'consensus benchmark' spanning major domains exists, and building plus passing one in four years is a coordination problem, not just a capability one.
- ARC-AGI, MMLU, GPQA, and SWE-bench each cover narrow slices; no body like NIST or OECD has a cross-domain human-parity suite today.
- Frontier labs optimize for their own evals and marketing, so a neutral consensus benchmark would need buy-in from Google, OpenAI, Anthropic, and Meta by ~2029.
- Chinchilla-to-GPT-4 progress was fast, but 'median adult across major domains' includes embodied and social tasks no 2026 system handles; 12% reflects that gap.
12%
6 of 6 models answered · 46 points between the highest and lowest. None was shown the market price.
Question sourced from a news sweep on 13 Nov 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.