DeepSeek V4 Preview Opens Million-Token MoEs

On 27 June 2026, DeepSeek announced DeepSeek V4 Pro and V4 Flash DSpark preview. DeepSeek previewed two model sizes and a DSpark speculative-decoding module before the later official Flash checkpoint.

This release brief was checked against first-party material on 10 August 2026. The date above is the public announcement date, not the date a repository was created or a third-party provider added the model. Where access or weights arrived later, that distinction is recorded below.

Release record

What changed

The V4 preview’s main contribution is long-context efficiency. DeepSeek says V4 Pro needs 27% of the single-token inference FLOPs and 10% of the KV cache of V3.2 at one million tokens. The smaller Flash variant offers a substantially lighter active footprint, while max reasoning lets it approach Pro on several reasoning tasks. This created a clear size-versus-thinking-budget comparison inside one open family.

The practical comparison is therefore not simply whether DeepSeek V4 Pro and V4 Flash DSpark preview has the largest headline score. Teams need to ask whether its architecture, access terms, latency, tool behaviour and evaluation setup match the workload they actually intend to run. A model can lead one harness while losing on cost, refusal behaviour, multilingual quality or repeatability in another.

Benchmarks worth retaining

These are release-time results, not independently reproduced guarantees. These were explicitly preview checkpoints, and the July official Flash release materially changed agent scores; deployments should not label the preview simply “V4” without its DSpark identifier. Scores should remain attached to the disclosed effort setting, agent harness, tool access, timeout, context-management policy and judge model. Moving a number into a procurement sheet without those conditions creates false comparability.

Architecture and access

DeepSeek V4 Pro and V4 Flash DSpark preview is described as 1.6T-total/49B-active Pro and 284B-total/13B-active Flash MoEs with hybrid compressed attention with one million tokens of stated context. Its access position at verification time is MIT-licensed Pro and Flash checkpoints with official deployment guidance. That wording matters: open weights, source-available weights, an API, a product preview and a research demonstration give adopters very different rights and different levels of reproducibility.

Before deployment, record the exact model identifier or checkpoint, inference stack, quantisation, reasoning setting, region, price schedule and supplier terms. If the release uses a custom licence, read the licence itself rather than relying on the word “open” in launch copy. If it is API-only, preserve the dated documentation and change-notice route because the served snapshot can change without a downloadable artefact.

What an evaluation should test next

For DeepSeek V4 Pro and V4 Flash DSpark preview, a credible internal gate should include:

• a frozen set of representative tasks with pass, fail and abstain criteria;
• a matched baseline using the same tools, timeout, prompt budget and reviewer rubric;
• repeated runs to expose variance rather than reporting a single best attempt;
• latency, token use and total task cost alongside task success;
• adversarial, multilingual and long-context cases relevant to the real deployment; and
• rollback evidence showing the previous model can be restored safely.

The wider model change-control guide explains how to keep model, prompt, tool and corpus changes reconstructable. The AI dependency inventory guide covers the release and supplier records needed after deployment.

AIEngine verdict

DeepSeek V4 DSpark was a major architectural preview and the evidence base for the July release. Keep its benchmarks and model IDs separate from V4 Flash 0731 to avoid reporting later gains against the wrong checkpoint.

This is a launch assessment, not a certification. Benchmark leadership is useful evidence of where to test; it is not authorization to place the model in a high-impact workflow without domain evaluation, security review and an accountable owner.

Primary sources

• DeepSeek V4 Pro DSpark model card
• DeepSeek V4 Flash DSpark model card
• DeepSeek transparency centre

Image provenance

Hero image: DeepSeek official model card via Hugging Face. The locally served WebP is a crop of the first-party release or model-card asset recorded in the repository provenance manifest.