Elon Musk just told his team coding dies by December.

“The AI just creates binary directly and can create a much more efficient binary than any compiler.”

One problem.

ClearerThinking analyzed 43 of Musk’s time-bound predictions across two decades. On-time delivery rate: 16.28%. He claims he aims for the 50th percentile. He misses by 3x. Professional traders on Polymarket now formally trade “Elon Time” as a discount factor, netting tens of thousands consistently betting No on any Musk deadline under six months.

Now the technical reality nobody in the replies is discussing.

SWE-bench Verified, updated February 5, 2026: Claude Opus 4.6 leads at 79.2%. Gemini 3 Flash at 76.2%. GPT 5.2 at 75.4%. Grok 4 sits below all three. Grok Code Fast scored 70.8% on its own internal harness. Musk says state-of-the-art in two to three months. The gap is not closing.

And “AI creates binary directly” is not even wrong. It is incoherent. GCC and LLVM encode sixty years of register allocation, instruction scheduling, loop vectorization, and cache-line optimization. No model anywhere has demonstrated even 5% efficiency gains over modern compilers on equivalent tasks. The claim requires AI to simultaneously outperform purpose-built systems across hardware architecture, memory hierarchy, and branch prediction. No benchmark supports this. None.

But here is what the 16.28% crowd will never understand.

The prediction does not need to be true. It needs to attract capital.

xAI merged with SpaceX on February 2. Colossus runs 200,000 GPUs scaling to 1.5 million. Annual funding sits at $20-30 billion. The restructuring landed February 11.

This is not a forecast. This is an IPO prospectus disguised as a prophecy.

The man who promises the most impossible future attracts the most patient money. And at that game, his accuracy rate is 100%.​​​​​​​​​​​​​​​​