DeepSeek vs Kimi hiring gap stems from compute divide: what are the key facts?

DeepSeek's hiring favors winners of ACM, ICPC and similar competition medals, leaning toward young undergraduates.

DeepSeek vs Kimi hiring gap stems from compute divide: what are the key facts?

Kimi's hiring emphasizes hands-on experience, testing candidates on model cheating detection and large-scale experimentation ability.

DeepSeek vs Kimi hiring gap stems from compute divide: what are the key facts?

DeepSeek has more abundant compute than Kimi, allowing new hires to produce no results for the first two or three months.

DeepSeek vs Kimi hiring gap stems from compute divide: what are the key facts?

Kimi is compute-constrained, requiring employees to quickly convert limited compute into model improvements.

DeepSeek vs Kimi hiring gap stems from compute divide: what are the key facts?

Neither company values traditional big-tech experience, believing old paradigms can become baggage.

Physix Frontier · News Briefing Card (enbrief · Oct 25, 2026)

DeepSeek vs Kimi hiring gap stems from compute divide

KEY FACTS

DeepSeek's hiring favors winners of ACM, ICPC and similar competition medals, leaning toward young undergraduates.
Kimi's hiring emphasizes hands-on experience, testing candidates on model cheating detection and large-scale experimentation ability.
DeepSeek has more abundant compute than Kimi, allowing new hires to produce no results for the first two or three months.
Kimi is compute-constrained, requiring employees to quickly convert limited compute into model improvements.
Neither company values traditional big-tech experience, believing old paradigms can become baggage.

PHYSIX OBSERVATION

The compute gap is directly shaping AI companies' talent strategies: those with plenty dare to raise geniuses and bet on the future, while those short on compute can only buy ready-made firepower. This reminds practitioners to check a company's compute reserves first, then match whether they themselves are a potential stock or a battle-tested operator, otherwise even the strongest may fail to adapt.

Source: enbrief original report ↗