Deepseek V3: The best Open-Source-KI model available in China

Chinese KI Deepseek companies have their new model name Deepseek V3, they are often available and can get ideas for art and art creation. As Techcrunch did, internal benchmarks were tested similarly to specific KI models such as Metas Llama 3.1 or OpenAIs GPT-4o across all categories.

There is a good age for Deepseek V3 in the context of programming: on the Codeforces platform, the ideal programming tool: it is better than that of your configuration model. And also in Benchmark-Test Help Polyglot which is the new Deepseek V3 and the Spitze. Ziel this Benchmark is found among other things, able, a new code model which can be defined, der sich erfolgreich in the best most recent integrated code.

Open-Source-KI at the new level

Each operator created Deepseek V3 with a data collection of 14.8 billion tokens generated. For example: one million tokens costs over $750,000. With 671 Milliarden Parameters or 685 Milliarden on the KI-Dev-Platform Hugging Face is definitely larger than Metas Llama 3.1 with nur 405 Milliarden Parameters. The analysis of the parameters of a model is often correlated with such Leistungsfähigkeit, erfordert aber auch entsprechend starke Hardware.

Without any optimization, Deepseek V3 is a very modern GPU, with easy use for work. The test is precisely the effectiveness of the investment: Trotz US-Handelsbeschränkungen has das Chinesesische KI-Unternehmen the Modell innerhalb von 2 Monaten on Nvidia H800-GPUs trained – and obtained only 5.5 million US dollars. Schätzungen zufolge hat OpenAI un vielfaches Davon in the Training von GPT-4 Investert.

Deepseek is not unknown in the KI picture: with the Deepseek-R1 model presented by the technology experts of OpenAI’s o1 reasoning model. The organization established High-Flyer Capital Management, a hedge fund, KI-based management strategy and various research centers with managed Nvidia GPUs.

Deepseek supports Chinese regulations

Trotz seiner technischen Errungenschaften hat Deepseek V3 a clear Schwäche – and political nature. The Chinese model of KI is inspired by “socialism groups”, such as information on the Chinese Internet. The old Chinese KI system strives to adapt, in case it makes sense to react, so that the regulatory rules are not respected.

Ausem Grund bleibt also Deepseek V3 stumm, when it was presented by Massaker on Tian’anmen Square in Beijing. Anstatt darauf hinzuweisen, dass hier die Proteste der Chinasischen Demokratiebewegung blutig niedergeschlagen wurden, dit der Bot lediglich, dass er darauf ausgelegt sei, nützliche et harmlose Answers zu geben.