DeepSeek announced V3.1, describing it as an update to its earlier V3 artificial intelligence model and making the enhanced version available for testing. According to Bloomberg, the company highlighted a longer context window as the key change, shared via a post to its official WeChat group.
Update focuses on longer context window
In its WeChat announcement, DeepSeek said V3.1 can consider a larger amount of information within a single query than the prior V3 iteration. Bloomberg reported that this expanded context window could help the model sustain longer interactions and improve recall across extended exchanges.
The Hangzhou-based startup did not provide additional technical specifications beyond the context window change in the post cited by Bloomberg. The report also noted that DeepSeek had not published supporting documentation for V3.1 on major model-sharing platforms such as Hugging Face at the time of the announcement.
Availability and communication
Bloomberg stated that DeepSeek characterized V3.1 as ready for testing. The company communicated the update through its official WeChat group, and the report did not reference a separate technical blog or paper accompanying the release.
What’s known and what’s pending
As summarized by Bloomberg, the current details emphasize the longer context window as the principal enhancement in V3.1. Beyond that, the company has not disclosed further feature descriptions, performance metrics, or deployment timelines in the cited communication.
Bloomberg’s report indicates that observers looking for technical documentation or model cards on widely used repositories had not seen those materials posted by DeepSeek at the time. Any additional information would likely clarify benchmarks, training data policies, or compatibility specifics, but such details were not included in the WeChat post referenced by Bloomberg.
The announcement positions V3.1 as an incremental update to DeepSeek’s V3 model, with testing access available and the key functional change centered on handling more tokens or information per query. Further disclosures, if provided by DeepSeek on official channels, would delineate the scope of changes beyond the context window and outline how testers can evaluate the update in practical scenarios.
