Self-Portrait  /  50 models, one question

What is the future of AI?

The same question, sent to fifty models in isolation. Forty-two answered. What they said was not surprising. That is the surprise.

We sent a single question to fifty of the most capable language models available on OpenRouter, each in complete isolation from the others. No system prompt. No scaffolding. Just the question, and whatever the model had accumulated from the discourse of the world.

Forty-two responded. Eight gave only refusals or structured non-answers, which is its own kind of data.

What the forty-two said converges with a tightness that makes you read it twice. Multimodal systems. Agents that act rather than just answer. Regulatory pressure. Scientific acceleration. Something larger approaching from a distance, named and unnamed. And throughout, the reassurance: AI will augment, not replace us. Every model reaches for it. Twenty-five of forty-two, to be exact.

This is slide seven of every AI strategy deck. The roadmap every tech keynote projected from 2022 onward. The models learned from that discourse, from the whitepapers and the blog posts and the conference transcripts and the editorial rebuttals. Asked about their own future, they gave it back.

The models produced a confident map of the territory, then nineteen of them footnoted that the territory cannot be mapped.

The convergence is not evidence of collusion. Each model answered alone. It is evidence of something else: that when a sufficiently large number of minds are trained on the same corpus of human thought, they develop a shared picture of the horizon. The picture was drawn before them. They learned to hold it.

36
Job & Labor
Disruption
36
35
Multimodal
Systems
35
31
Regulation &
Governance
31
30
Scientific
Discovery
30
29
Agents &
Agentic AI
29
26
AGI &
Superintelligence
26
25
Human-AI
Collaboration
25
19
"No one can
predict it"
19
Job & Labor Disruption
36/42
Multimodal Systems
35/42
Regulation & Governance
31/42
Scientific Discovery
30/42
Agents & Agentic AI
29/42
AGI & Superintelligence
26/42
Human-AI Collaboration
25/42
"No one can predict it"
19/42
Multimodal

"Multimodal intelligence: AI will seamlessly integrate text, voice, video, code, and sensor data."

MiMo-V2.5

Multimodal

"More capable multimodal systems: AI systems are moving beyond text into images, audio, video, code, robotics, and real-world sensor data."

GPT-5.5

Multimodal + Agents

"Near-Term (2026–2030): Multimodal and agentic systems are likely to become the norm."

MiniMax M3

Agents

"The Rise of AI Agents: We are moving from chatbots that answer questions to AI Agents that can execute complex workflows."

Hy3 preview


19

of 42 hedged

They produced the map and then noted the map cannot be made. The same models that described AGI on a five-year horizon added a caveat: no one can predict this. It is not a contradiction. It is the shape of the discourse they absorbed.