Model

DeepSeek V5

A next-generation open-source large language model from DeepSeek. Reported specifications include a parameter scale of approximately 3 trillion and support for a context window of up to 1 million tokens. The model is positioned to compete with leading closed-source models.

Timeline (last 90 days)

Nov 4

DeepSeek V5 will use Huawei Ascend 950 as its core training compute for the first time, with implementation expected next year.

Unverified50%

Oct 28

Three model names—V5 Open, V5, and V5 Next—appear in the DeepSeek Harness code, and the relevant identifiers are also written into the test files.

Unverified50%

Sep 24

DeepSeek is preparing its next-generation large model DeepSeek V5, and rumors circulating suggest the model will be benchmarked against GPT-6 Astra and attempt to match the performance of Fable 5.1.

Expired50%

All Facts (2)

Unverified

Three model names—V5 Open, V5, and V5 Next—appear in the DeepSeek Harness code, and the relevant identifiers are also written into the test files.

50%Unverified

DeepSeek V5 will use Huawei Ascend 950 as its core training compute for the first time, with implementation expected next year.

50%

Source Articles

GPT-6.2 Astra Leaked: Low-End Config Beats Previous Top-Tier, SVG Generation StunsNov 4 DeepSeek V5 Reportedly Moves to Huawei Ascend for Training as GPT-6.2 and Qwen 4.0 SurfaceNov 4 DeepSeek V5 Leaked: Open-Source Model Takes Aim at Astra as Multi-Front Model Battle LoomsSep 24