DeepSeek V5: Code traces and rumors, but no official date

There is still no official V5 from the Chinese AI company DeepSeek. As of the end of October 2026, the company had not published an announcement, a model card (a technical description of the model), a name for developers or a release date. Everything about the timing is therefore based on clues and rumors.

Traces in the code

A Chinese analysis from AI-Indeed points to three placeholders that were added on September 24, 2026, to a test file in DeepSeek's Harness repository (a place where the company's code is stored): deepseek-v5, deepseek-v5-open and deepseek-v5-next. They are currently inactive.

According to the analysis, deepseek-v5-open sets minimal requirements for the client version, so older applications can gain access through server settings. Deepseek-v5-next points to version 99.0.0-alpha.1, which suggests it requires updated applications.

The analysis judges that November 2026 is the most likely timing. The reasoning is that the Flash version, according to the analysis, arrived on September 17 and needs distance from a Pro release, and that a month of technical preparation has passed. According to the analysis, however, V5 could also arrive earlier or as a preview, like V4. The placeholders are not proof that a release is imminent.

The July 24 misunderstanding

One date that has circulated is July 24. According to the blogger Thomas Wiegold, it is based on a misunderstanding. That day, DeepSeek retired the old model names deepseek-chat and deepseek-reasoner, which pointed to V4 Flash. It was routine maintenance and not a new model. Wiegold believes that anyone who gives a fixed V5 date is guessing.

For comparison, V4 came out on April 24, 2026, and in June DeepSeek closed its first external funding round of roughly $7.4 billion.

Rumor about Huawei chip

An article from Phoenix Net on November 3 cites anonymous industry insiders, quoted via Financial Union. According to them, V5 is to be trained primarily on Huawei's Ascend 950 chip, and a rollout is expected in 2026. That would make it DeepSeek's first training run that natively depends on the chip.

The article also says DeepSeek is planning a data center in Ulanqab for Ascend 950 resources, and that code based on CUDA (Nvidia's programming platform) has been moved to open Python environments to work across chip architectures. Huawei has not invested in DeepSeek. However, the information comes from anonymous sources via an intermediary.

In the AI community, according to the analysis, V5 is expected to have stronger reasoning, coding and agent capabilities than V4 and to be cheaper to run. These are expectations, not something DeepSeek has promised.